Stuff

* Fix

* Fix greptile

* Fix validation

* Fix comments

* Lint

* Fix

* remove passed in workspace id ref

* Fix comments

---------

Co-authored-by: Vikhyath Mondreti <vikhyath@simstudio.ai>

README.md

@@ -172,31 +172,6 @@ Key environment variables for self-hosted deployments. See [`.env.example`](apps
 | `API_ENCRYPTION_KEY` | Yes | Encrypts API keys (`openssl rand -hex 32`) |
 | `COPILOT_API_KEY` | No | API key from sim.ai for Copilot features |
 
-## Troubleshooting
-
-### Ollama models not showing in dropdown (Docker)
-
-If you're running Ollama on your host machine and Sim in Docker, change `OLLAMA_URL` from `localhost` to `host.docker.internal`:
-
-```bash
-OLLAMA_URL=http://host.docker.internal:11434 docker compose -f docker-compose.prod.yml up -d
-```
-
-See [Using an External Ollama Instance](#using-an-external-ollama-instance) for details.
-
-### Database connection issues
-
-Ensure PostgreSQL has the pgvector extension installed. When using Docker, wait for the database to be healthy before running migrations.
-
-### Port conflicts
-
-If ports 3000, 3002, or 5432 are in use, configure alternatives:
-
-```bash
-# Custom ports
-NEXT_PUBLIC_APP_URL=http://localhost:3100 POSTGRES_PORT=5433 docker compose up -d
-```
-
 ## Tech Stack
 
 - **Framework**: [Next.js](https://nextjs.org/) (App Router)

apps/sim/app/api/copilot/chat/route.ts

@@ -1,13 +1,13 @@
 import { db } from '@sim/db'
-import { copilotChats } from '@sim/db/schema'
+import { copilotChats, permissions, workflow } from '@sim/db/schema'
 import { createLogger } from '@sim/logger'
-import { and, desc, eq } from 'drizzle-orm'
+import { and, asc, desc, eq, inArray, or } from 'drizzle-orm'
 import { type NextRequest, NextResponse } from 'next/server'
 import { z } from 'zod'
 import { getSession } from '@/lib/auth'
 import { generateChatTitle } from '@/lib/copilot/chat-title'
 import { getCopilotModel } from '@/lib/copilot/config'
-import { SIM_AGENT_API_URL_DEFAULT, SIM_AGENT_VERSION } from '@/lib/copilot/constants'
+import { SIM_AGENT_VERSION } from '@/lib/copilot/constants'
 import { COPILOT_MODEL_IDS, COPILOT_REQUEST_MODES } from '@/lib/copilot/models'
 import {
   authenticateCopilotRequestSessionOnly,
@@ -23,10 +23,10 @@ import { CopilotFiles } from '@/lib/uploads'
 import { createFileContent } from '@/lib/uploads/utils/file-utils'
 import { tools } from '@/tools/registry'
 import { getLatestVersionTools, stripVersionSuffix } from '@/tools/utils'
+import { orchestrateCopilotStream } from '@/lib/copilot/orchestrator'
 
 const logger = createLogger('CopilotChatAPI')
 
-const SIM_AGENT_API_URL = env.SIM_AGENT_API_URL || SIM_AGENT_API_URL_DEFAULT
 
 const FileAttachmentSchema = z.object({
   id: z.string(),
@@ -40,7 +40,8 @@ const ChatMessageSchema = z.object({
   message: z.string().min(1, 'Message is required'),
   userMessageId: z.string().optional(), // ID from frontend for the user message
   chatId: z.string().optional(),
-  workflowId: z.string().min(1, 'Workflow ID is required'),
+  workflowId: z.string().optional(),
+  workflowName: z.string().optional(),
   model: z.enum(COPILOT_MODEL_IDS).optional().default('claude-4.5-opus'),
   mode: z.enum(COPILOT_REQUEST_MODES).optional().default('agent'),
   prefetch: z.boolean().optional(),
@@ -78,6 +79,54 @@ const ChatMessageSchema = z.object({
   commands: z.array(z.string()).optional(),
 })
 
+async function resolveWorkflowId(
+  userId: string,
+  workflowId?: string,
+  workflowName?: string
+): Promise<{ workflowId: string; workflowName?: string } | null> {
+  // If workflowId provided, use it directly
+  if (workflowId) {
+    return { workflowId }
+  }
+
+  // Get user's accessible workflows
+  const workspaceIds = await db
+    .select({ entityId: permissions.entityId })
+    .from(permissions)
+    .where(and(eq(permissions.userId, userId), eq(permissions.entityType, 'workspace')))
+
+  const workspaceIdList = workspaceIds.map((row) => row.entityId)
+
+  const workflowConditions = [eq(workflow.userId, userId)]
+  if (workspaceIdList.length > 0) {
+    workflowConditions.push(inArray(workflow.workspaceId, workspaceIdList))
+  }
+
+  const workflows = await db
+    .select()
+    .from(workflow)
+    .where(or(...workflowConditions))
+    .orderBy(asc(workflow.sortOrder), asc(workflow.createdAt), asc(workflow.id))
+
+  if (workflows.length === 0) {
+    return null
+  }
+
+  // If workflowName provided, find matching workflow
+  if (workflowName) {
+    const match = workflows.find(
+      (w) => String(w.name || '').trim().toLowerCase() === workflowName.toLowerCase()
+    )
+    if (match) {
+      return { workflowId: match.id, workflowName: match.name || undefined }
+    }
+    return null
+  }
+
+  // Default to first workflow
+  return { workflowId: workflows[0].id, workflowName: workflows[0].name || undefined }
+}
+
 /**
  * POST /api/copilot/chat
  * Send messages to sim agent and handle chat persistence
@@ -100,7 +149,8 @@ export async function POST(req: NextRequest) {
       message,
       userMessageId,
       chatId,
-      workflowId,
+      workflowId: providedWorkflowId,
+      workflowName,
       model,
       mode,
       prefetch,
@@ -113,6 +163,16 @@ export async function POST(req: NextRequest) {
       contexts,
       commands,
     } = ChatMessageSchema.parse(body)
+
+    // Resolve workflowId - if not provided, use first workflow or find by name
+    const resolved = await resolveWorkflowId(authenticatedUserId, providedWorkflowId, workflowName)
+    if (!resolved) {
+      return createBadRequestResponse(
+        'No workflows found. Create a workflow first or provide a valid workflowId.'
+      )
+    }
+    const workflowId = resolved.workflowId
+
     // Ensure we have a consistent user message ID for this request
     const userMessageIdToUse = userMessageId || crypto.randomUUID()
     try {
@@ -465,77 +525,19 @@ export async function POST(req: NextRequest) {
       })
     } catch {}
 
-    const simAgentResponse = await fetch(`${SIM_AGENT_API_URL}/api/chat-completion-streaming`, {
-      method: 'POST',
-      headers: {
-        'Content-Type': 'application/json',
-        ...(env.COPILOT_API_KEY ? { 'x-api-key': env.COPILOT_API_KEY } : {}),
-      },
-      body: JSON.stringify(requestPayload),
-    })
-
-    if (!simAgentResponse.ok) {
-      if (simAgentResponse.status === 401 || simAgentResponse.status === 402) {
-        // Rethrow status only; client will render appropriate assistant message
-        return new NextResponse(null, { status: simAgentResponse.status })
-      }
-
-      const errorText = await simAgentResponse.text().catch(() => '')
-      logger.error(`[${tracker.requestId}] Sim agent API error:`, {
-        status: simAgentResponse.status,
-        error: errorText,
-      })
-
-      return NextResponse.json(
-        { error: `Sim agent API error: ${simAgentResponse.statusText}` },
-        { status: simAgentResponse.status }
-      )
-    }
-
-    // If streaming is requested, forward the stream and update chat later
-    if (stream && simAgentResponse.body) {
-      // Create user message to save
-      const userMessage = {
-        id: userMessageIdToUse, // Consistent ID used for request and persistence
-        role: 'user',
-        content: message,
-        timestamp: new Date().toISOString(),
-        ...(fileAttachments && fileAttachments.length > 0 && { fileAttachments }),
-        ...(Array.isArray(contexts) && contexts.length > 0 && { contexts }),
-        ...(Array.isArray(contexts) &&
-          contexts.length > 0 && {
-            contentBlocks: [{ type: 'contexts', contexts: contexts as any, timestamp: Date.now() }],
-          }),
-      }
-
-      // Create a pass-through stream that captures the response
+    if (stream) {
       const transformedStream = new ReadableStream({
         async start(controller) {
           const encoder = new TextEncoder()
-          let assistantContent = ''
-          const toolCalls: any[] = []
-          let buffer = ''
-          const isFirstDone = true
-          let responseIdFromStart: string | undefined
-          let responseIdFromDone: string | undefined
-          // Track tool call progress to identify a safe done event
-          const announcedToolCallIds = new Set<string>()
-          const startedToolExecutionIds = new Set<string>()
-          const completedToolExecutionIds = new Set<string>()
-          let lastDoneResponseId: string | undefined
-          let lastSafeDoneResponseId: string | undefined
 
-          // Send chatId as first event
           if (actualChatId) {
-            const chatIdEvent = `data: ${JSON.stringify({
-              type: 'chat_id',
-              chatId: actualChatId,
-            })}\n\n`
-            controller.enqueue(encoder.encode(chatIdEvent))
-            logger.debug(`[${tracker.requestId}] Sent initial chatId event to client`)
+            controller.enqueue(
+              encoder.encode(
+                `data: ${JSON.stringify({ type: 'chat_id', chatId: actualChatId })}\n\n`
+              )
+            )
           }
 
-          // Start title generation in parallel if needed
           if (actualChatId && !currentChat?.title && conversationHistory.length === 0) {
             generateChatTitle(message)
               .then(async (title) => {
@@ -547,311 +549,61 @@ export async function POST(req: NextRequest) {
                       updatedAt: new Date(),
                     })
                     .where(eq(copilotChats.id, actualChatId!))
-
-                  const titleEvent = `data: ${JSON.stringify({
-                    type: 'title_updated',
-                    title: title,
-                  })}\n\n`
-                  controller.enqueue(encoder.encode(titleEvent))
-                  logger.info(`[${tracker.requestId}] Generated and saved title: ${title}`)
+                  controller.enqueue(
+                    encoder.encode(`data: ${JSON.stringify({ type: 'title_updated', title })}\n\n`)
+                  )
                 }
               })
               .catch((error) => {
                 logger.error(`[${tracker.requestId}] Title generation failed:`, error)
               })
-          } else {
-            logger.debug(`[${tracker.requestId}] Skipping title generation`)
           }
 
-          // Forward the sim agent stream and capture assistant response
-          const reader = simAgentResponse.body!.getReader()
-          const decoder = new TextDecoder()
-
           try {
-            while (true) {
-              const { done, value } = await reader.read()
-              if (done) {
-                break
-              }
-
-              // Decode and parse SSE events for logging and capturing content
-              const decodedChunk = decoder.decode(value, { stream: true })
-              buffer += decodedChunk
-
-              const lines = buffer.split('\n')
-              buffer = lines.pop() || '' // Keep incomplete line in buffer
-
-              for (const line of lines) {
-                if (line.trim() === '') continue // Skip empty lines
-
-                if (line.startsWith('data: ') && line.length > 6) {
-                  try {
-                    const jsonStr = line.slice(6)
-
-                    // Check if the JSON string is unusually large (potential streaming issue)
-                    if (jsonStr.length > 50000) {
-                      // 50KB limit
-                      logger.warn(`[${tracker.requestId}] Large SSE event detected`, {
-                        size: jsonStr.length,
-                        preview: `${jsonStr.substring(0, 100)}...`,
-                      })
-                    }
-
-                    const event = JSON.parse(jsonStr)
-
-                    // Log different event types comprehensively
-                    switch (event.type) {
-                      case 'content':
-                        if (event.data) {
-                          assistantContent += event.data
-                        }
-                        break
-
-                      case 'reasoning':
-                        logger.debug(
-                          `[${tracker.requestId}] Reasoning chunk received (${(event.data || event.content || '').length} chars)`
-                        )
-                        break
-
-                      case 'tool_call':
-                        if (!event.data?.partial) {
-                          toolCalls.push(event.data)
-                          if (event.data?.id) {
-                            announcedToolCallIds.add(event.data.id)
-                          }
-                        }
-                        break
-
-                      case 'tool_generating':
-                        if (event.toolCallId) {
-                          startedToolExecutionIds.add(event.toolCallId)
-                        }
-                        break
-
-                      case 'tool_result':
-                        if (event.toolCallId) {
-                          completedToolExecutionIds.add(event.toolCallId)
-                        }
-                        break
-
-                      case 'tool_error':
-                        logger.error(`[${tracker.requestId}] Tool error:`, {
-                          toolCallId: event.toolCallId,
-                          toolName: event.toolName,
-                          error: event.error,
-                          success: event.success,
-                        })
-                        if (event.toolCallId) {
-                          completedToolExecutionIds.add(event.toolCallId)
-                        }
-                        break
-
-                      case 'start':
-                        if (event.data?.responseId) {
-                          responseIdFromStart = event.data.responseId
-                        }
-                        break
-
-                      case 'done':
-                        if (event.data?.responseId) {
-                          responseIdFromDone = event.data.responseId
-                          lastDoneResponseId = responseIdFromDone
-
-                          // Mark this done as safe only if no tool call is currently in progress or pending
-                          const announced = announcedToolCallIds.size
-                          const completed = completedToolExecutionIds.size
-                          const started = startedToolExecutionIds.size
-                          const hasToolInProgress = announced > completed || started > completed
-                          if (!hasToolInProgress) {
-                            lastSafeDoneResponseId = responseIdFromDone
-                          }
-                        }
-                        break
-
-                      case 'error':
-                        break
-
-                      default:
-                    }
-
-                    // Emit to client: rewrite 'error' events into user-friendly assistant message
-                    if (event?.type === 'error') {
-                      try {
-                        const displayMessage: string =
-                          (event?.data && (event.data.displayMessage as string)) ||
-                          'Sorry, I encountered an error. Please try again.'
-                        const formatted = `_${displayMessage}_`
-                        // Accumulate so it persists to DB as assistant content
-                        assistantContent += formatted
-                        // Send as content chunk
-                        try {
-                          controller.enqueue(
-                            encoder.encode(
-                              `data: ${JSON.stringify({ type: 'content', data: formatted })}\n\n`
-                            )
-                          )
-                        } catch (enqueueErr) {
-                          reader.cancel()
-                          break
-                        }
-                        // Then close this response cleanly for the client
-                        try {
-                          controller.enqueue(
-                            encoder.encode(`data: ${JSON.stringify({ type: 'done' })}\n\n`)
-                          )
-                        } catch (enqueueErr) {
-                          reader.cancel()
-                          break
-                        }
-                      } catch {}
-                      // Do not forward the original error event
-                    } else {
-                      // Forward original event to client
-                      try {
-                        controller.enqueue(encoder.encode(`data: ${jsonStr}\n\n`))
-                      } catch (enqueueErr) {
-                        reader.cancel()
-                        break
-                      }
-                    }
-                  } catch (e) {
-                    // Enhanced error handling for large payloads and parsing issues
-                    const lineLength = line.length
-                    const isLargePayload = lineLength > 10000
-
-                    if (isLargePayload) {
-                      logger.error(
-                        `[${tracker.requestId}] Failed to parse large SSE event (${lineLength} chars)`,
-                        {
-                          error: e,
-                          preview: `${line.substring(0, 200)}...`,
-                          size: lineLength,
-                        }
-                      )
-                    } else {
-                      logger.warn(
-                        `[${tracker.requestId}] Failed to parse SSE event: "${line.substring(0, 200)}..."`,
-                        e
-                      )
-                    }
-                  }
-                } else if (line.trim() && line !== 'data: [DONE]') {
-                  logger.debug(`[${tracker.requestId}] Non-SSE line from sim agent: "${line}"`)
-                }
-              }
-            }
-
-            // Process any remaining buffer
-            if (buffer.trim()) {
-              logger.debug(`[${tracker.requestId}] Processing remaining buffer: "${buffer}"`)
-              if (buffer.startsWith('data: ')) {
+            const result = await orchestrateCopilotStream(requestPayload, {
+              userId: authenticatedUserId,
+              workflowId,
+              chatId: actualChatId,
+              autoExecuteTools: true,
+              interactive: true,
+              onEvent: async (event) => {
                 try {
-                  const jsonStr = buffer.slice(6)
-                  const event = JSON.parse(jsonStr)
-                  if (event.type === 'content' && event.data) {
-                    assistantContent += event.data
-                  }
-                  // Forward remaining event, applying same error rewrite behavior
-                  if (event?.type === 'error') {
-                    const displayMessage: string =
-                      (event?.data && (event.data.displayMessage as string)) ||
-                      'Sorry, I encountered an error. Please try again.'
-                    const formatted = `_${displayMessage}_`
-                    assistantContent += formatted
-                    try {
-                      controller.enqueue(
-                        encoder.encode(
-                          `data: ${JSON.stringify({ type: 'content', data: formatted })}\n\n`
-                        )
-                      )
-                      controller.enqueue(
-                        encoder.encode(`data: ${JSON.stringify({ type: 'done' })}\n\n`)
-                      )
-                    } catch (enqueueErr) {
-                      reader.cancel()
-                    }
-                  } else {
-                    try {
-                      controller.enqueue(encoder.encode(`data: ${jsonStr}\n\n`))
-                    } catch (enqueueErr) {
-                      reader.cancel()
-                    }
-                  }
-                } catch (e) {
-                  logger.warn(`[${tracker.requestId}] Failed to parse final buffer: "${buffer}"`)
+                  controller.enqueue(encoder.encode(`data: ${JSON.stringify(event)}\n\n`))
+                } catch {
+                  controller.error('Failed to forward SSE event')
                 }
-              }
-            }
-
-            // Log final streaming summary
-            logger.info(`[${tracker.requestId}] Streaming complete summary:`, {
-              totalContentLength: assistantContent.length,
-              toolCallsCount: toolCalls.length,
-              hasContent: assistantContent.length > 0,
-              toolNames: toolCalls.map((tc) => tc?.name).filter(Boolean),
+              },
             })
 
-            // NOTE: Messages are saved by the client via update-messages endpoint with full contentBlocks.
-            // Server only updates conversationId here to avoid overwriting client's richer save.
-            if (currentChat) {
-              // Persist only a safe conversationId to avoid continuing from a state that expects tool outputs
-              const previousConversationId = currentChat?.conversationId as string | undefined
-              const responseId = lastSafeDoneResponseId || previousConversationId || undefined
-
-              if (responseId) {
-                await db
-                  .update(copilotChats)
-                  .set({
-                    updatedAt: new Date(),
-                    conversationId: responseId,
-                  })
-                  .where(eq(copilotChats.id, actualChatId!))
-
-                logger.info(
-                  `[${tracker.requestId}] Updated conversationId for chat ${actualChatId}`,
-                  {
-                    updatedConversationId: responseId,
-                  }
-                )
-              }
+            if (currentChat && result.conversationId) {
+              await db
+                .update(copilotChats)
+                .set({
+                  updatedAt: new Date(),
+                  conversationId: result.conversationId,
+                })
+                .where(eq(copilotChats.id, actualChatId!))
             }
           } catch (error) {
-            logger.error(`[${tracker.requestId}] Error processing stream:`, error)
-
-            // Send an error event to the client before closing so it knows what happened
-            try {
-              const errorMessage =
-                error instanceof Error && error.message === 'terminated'
-                  ? 'Connection to AI service was interrupted. Please try again.'
-                  : 'An unexpected error occurred while processing the response.'
-              const encoder = new TextEncoder()
-
-              // Send error as content so it shows in the chat
-              controller.enqueue(
-                encoder.encode(
-                  `data: ${JSON.stringify({ type: 'content', data: `\n\n_${errorMessage}_` })}\n\n`
-                )
+            logger.error(`[${tracker.requestId}] Orchestration error:`, error)
+            controller.enqueue(
+              encoder.encode(
+                `data: ${JSON.stringify({
+                  type: 'error',
+                  data: {
+                    displayMessage:
+                      'An unexpected error occurred while processing the response.',
+                  },
+                })}\n\n`
               )
-              // Send done event to properly close the stream on client
-              controller.enqueue(encoder.encode(`data: ${JSON.stringify({ type: 'done' })}\n\n`))
-            } catch (enqueueError) {
-              // Stream might already be closed, that's ok
-              logger.warn(
-                `[${tracker.requestId}] Could not send error event to client:`,
-                enqueueError
-              )
-            }
+            )
           } finally {
-            try {
-              controller.close()
-            } catch {
-              // Controller might already be closed
-            }
+            controller.close()
           }
         },
       })
 
-      const response = new Response(transformedStream, {
+      return new Response(transformedStream, {
         headers: {
           'Content-Type': 'text/event-stream',
           'Cache-Control': 'no-cache',
@@ -859,43 +611,31 @@ export async function POST(req: NextRequest) {
           'X-Accel-Buffering': 'no',
         },
       })
-
-      logger.info(`[${tracker.requestId}] Returning streaming response to client`, {
-        duration: tracker.getDuration(),
-        chatId: actualChatId,
-        headers: {
-          'Content-Type': 'text/event-stream',
-          'Cache-Control': 'no-cache',
-          Connection: 'keep-alive',
-        },
-      })
-
-      return response
     }
 
-    // For non-streaming responses
-    const responseData = await simAgentResponse.json()
-    logger.info(`[${tracker.requestId}] Non-streaming response from sim agent:`, {
+    const nonStreamingResult = await orchestrateCopilotStream(requestPayload, {
+      userId: authenticatedUserId,
+      workflowId,
+      chatId: actualChatId,
+      autoExecuteTools: true,
+      interactive: true,
+    })
+
+    const responseData = {
+      content: nonStreamingResult.content,
+      toolCalls: nonStreamingResult.toolCalls,
+      model: selectedModel,
+      provider: providerConfig?.provider || env.COPILOT_PROVIDER || 'openai',
+    }
+
+    logger.info(`[${tracker.requestId}] Non-streaming response from orchestrator:`, {
       hasContent: !!responseData.content,
       contentLength: responseData.content?.length || 0,
       model: responseData.model,
       provider: responseData.provider,
       toolCallsCount: responseData.toolCalls?.length || 0,
-      hasTokens: !!responseData.tokens,
     })
 
-    // Log tool calls if present
-    if (responseData.toolCalls?.length > 0) {
-      responseData.toolCalls.forEach((toolCall: any) => {
-        logger.info(`[${tracker.requestId}] Tool call in response:`, {
-          id: toolCall.id,
-          name: toolCall.name,
-          success: toolCall.success,
-          result: `${JSON.stringify(toolCall.result).substring(0, 200)}...`,
-        })
-      })
-    }
-
     // Save messages if we have a chat
     if (currentChat && responseData.content) {
       const userMessage = {
@@ -947,6 +687,9 @@ export async function POST(req: NextRequest) {
         .set({
           messages: updatedMessages,
           updatedAt: new Date(),
+          ...(nonStreamingResult.conversationId
+            ? { conversationId: nonStreamingResult.conversationId }
+            : {}),
         })
         .where(eq(copilotChats.id, actualChatId!))
     }

apps/sim/app/api/v1/copilot/chat/route.ts

@@ -0,0 +1,157 @@
+import { db } from '@sim/db'
+import { permissions, workflow } from '@sim/db/schema'
+import { createLogger } from '@sim/logger'
+import { and, asc, eq, inArray, or } from 'drizzle-orm'
+import { type NextRequest, NextResponse } from 'next/server'
+import { z } from 'zod'
+import { authenticateV1Request } from '@/app/api/v1/auth'
+import { getCopilotModel } from '@/lib/copilot/config'
+import { SIM_AGENT_VERSION } from '@/lib/copilot/constants'
+import { COPILOT_REQUEST_MODES } from '@/lib/copilot/models'
+import { orchestrateCopilotStream } from '@/lib/copilot/orchestrator'
+
+const logger = createLogger('CopilotHeadlessAPI')
+
+const RequestSchema = z.object({
+  message: z.string().min(1, 'message is required'),
+  workflowId: z.string().optional(),
+  workflowName: z.string().optional(),
+  chatId: z.string().optional(),
+  mode: z.enum(COPILOT_REQUEST_MODES).optional().default('agent'),
+  model: z.string().optional(),
+  autoExecuteTools: z.boolean().optional().default(true),
+  timeout: z.number().optional().default(300000),
+})
+
+async function resolveWorkflowId(
+  userId: string,
+  workflowId?: string,
+  workflowName?: string
+): Promise<{ workflowId: string; workflowName?: string } | null> {
+  // If workflowId provided, use it directly
+  if (workflowId) {
+    return { workflowId }
+  }
+
+  // Get user's accessible workflows
+  const workspaceIds = await db
+    .select({ entityId: permissions.entityId })
+    .from(permissions)
+    .where(and(eq(permissions.userId, userId), eq(permissions.entityType, 'workspace')))
+
+  const workspaceIdList = workspaceIds.map((row) => row.entityId)
+
+  const workflowConditions = [eq(workflow.userId, userId)]
+  if (workspaceIdList.length > 0) {
+    workflowConditions.push(inArray(workflow.workspaceId, workspaceIdList))
+  }
+
+  const workflows = await db
+    .select()
+    .from(workflow)
+    .where(or(...workflowConditions))
+    .orderBy(asc(workflow.sortOrder), asc(workflow.createdAt), asc(workflow.id))
+
+  if (workflows.length === 0) {
+    return null
+  }
+
+  // If workflowName provided, find matching workflow
+  if (workflowName) {
+    const match = workflows.find(
+      (w) => String(w.name || '').trim().toLowerCase() === workflowName.toLowerCase()
+    )
+    if (match) {
+      return { workflowId: match.id, workflowName: match.name || undefined }
+    }
+    return null
+  }
+
+  // Default to first workflow
+  return { workflowId: workflows[0].id, workflowName: workflows[0].name || undefined }
+}
+
+/**
+ * POST /api/v1/copilot/chat
+ * Headless copilot endpoint for server-side orchestration.
+ * 
+ * workflowId is optional - if not provided:
+ * - If workflowName is provided, finds that workflow
+ * - Otherwise uses the user's first workflow as context
+ * - The copilot can still operate on any workflow using list_user_workflows
+ */
+export async function POST(req: NextRequest) {
+  const auth = await authenticateV1Request(req)
+  if (!auth.authenticated || !auth.userId) {
+    return NextResponse.json({ success: false, error: auth.error || 'Unauthorized' }, { status: 401 })
+  }
+
+  try {
+    const body = await req.json()
+    const parsed = RequestSchema.parse(body)
+    const defaults = getCopilotModel('chat')
+    const selectedModel = parsed.model || defaults.model
+
+    // Resolve workflow ID
+    const resolved = await resolveWorkflowId(auth.userId, parsed.workflowId, parsed.workflowName)
+    if (!resolved) {
+      return NextResponse.json(
+        { success: false, error: 'No workflows found. Create a workflow first or provide a valid workflowId.' },
+        { status: 400 }
+      )
+    }
+
+    // Transform mode to transport mode (same as client API)
+    // build and agent both map to 'agent' on the backend
+    const effectiveMode = parsed.mode === 'agent' ? 'build' : parsed.mode
+    const transportMode = effectiveMode === 'build' ? 'agent' : effectiveMode
+
+    // Always generate a chatId - required for artifacts system to work with subagents
+    const chatId = parsed.chatId || crypto.randomUUID()
+
+    const requestPayload = {
+      message: parsed.message,
+      workflowId: resolved.workflowId,
+      userId: auth.userId,
+      stream: true,
+      streamToolCalls: true,
+      model: selectedModel,
+      mode: transportMode,
+      messageId: crypto.randomUUID(),
+      version: SIM_AGENT_VERSION,
+      headless: true, // Enable cross-workflow operations via workflowId params
+      chatId,
+    }
+
+    const result = await orchestrateCopilotStream(requestPayload, {
+      userId: auth.userId,
+      workflowId: resolved.workflowId,
+      chatId,
+      autoExecuteTools: parsed.autoExecuteTools,
+      timeout: parsed.timeout,
+      interactive: false,
+    })
+
+    return NextResponse.json({
+      success: result.success,
+      content: result.content,
+      toolCalls: result.toolCalls,
+      chatId: result.chatId || chatId, // Return the chatId for conversation continuity
+      conversationId: result.conversationId,
+      error: result.error,
+    })
+  } catch (error) {
+    if (error instanceof z.ZodError) {
+      return NextResponse.json(
+        { success: false, error: 'Invalid request', details: error.errors },
+        { status: 400 }
+      )
+    }
+
+    logger.error('Headless copilot request failed', {
+      error: error instanceof Error ? error.message : String(error),
+    })
+    return NextResponse.json({ success: false, error: 'Internal server error' }, { status: 500 })
+  }
+}
+

apps/sim/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/copilot/components/tool-call/tool-call.tsx

@@ -1,5 +1,6 @@
 'use client'
 
+import { createLogger } from '@sim/logger'
 import { memo, useEffect, useMemo, useRef, useState } from 'react'
 import clsx from 'clsx'
 import { ChevronUp, LayoutList } from 'lucide-react'
@@ -25,6 +26,7 @@ import { getBlock } from '@/blocks/registry'
 import type { CopilotToolCall } from '@/stores/panel'
 import { useCopilotStore } from '@/stores/panel'
 import { CLASS_TOOL_METADATA } from '@/stores/panel/copilot/store'
+import { COPILOT_SERVER_ORCHESTRATED } from '@/lib/copilot/orchestrator/config'
 import type { SubAgentContentBlock } from '@/stores/panel/copilot/types'
 import { useWorkflowStore } from '@/stores/workflows/workflow/store'
 
@@ -1259,12 +1261,36 @@ function shouldShowRunSkipButtons(toolCall: CopilotToolCall): boolean {
   return false
 }
 
+const toolCallLogger = createLogger('CopilotToolCall')
+
+async function sendToolDecision(toolCallId: string, status: 'accepted' | 'rejected') {
+  try {
+    await fetch('/api/copilot/confirm', {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({ toolCallId, status }),
+    })
+  } catch (error) {
+    toolCallLogger.warn('Failed to send tool decision', {
+      toolCallId,
+      status,
+      error: error instanceof Error ? error.message : String(error),
+    })
+  }
+}
+
 async function handleRun(
   toolCall: CopilotToolCall,
   setToolCallState: any,
   onStateChange?: any,
   editedParams?: any
 ) {
+  if (COPILOT_SERVER_ORCHESTRATED) {
+    setToolCallState(toolCall, 'executing')
+    onStateChange?.('executing')
+    await sendToolDecision(toolCall.id, 'accepted')
+    return
+  }
   const instance = getClientTool(toolCall.id)
 
   if (!instance && isIntegrationTool(toolCall.name)) {
@@ -1309,6 +1335,12 @@ async function handleRun(
 }
 
 async function handleSkip(toolCall: CopilotToolCall, setToolCallState: any, onStateChange?: any) {
+  if (COPILOT_SERVER_ORCHESTRATED) {
+    setToolCallState(toolCall, 'rejected')
+    onStateChange?.('rejected')
+    await sendToolDecision(toolCall.id, 'rejected')
+    return
+  }
   const instance = getClientTool(toolCall.id)
 
   if (!instance && isIntegrationTool(toolCall.name)) {

apps/sim/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/credential-selector/credential-selector.tsx

@@ -1,6 +1,6 @@
 'use client'
 
-import { useCallback, useEffect, useMemo, useState } from 'react'
+import { createElement, useCallback, useEffect, useMemo, useState } from 'react'
 import { createLogger } from '@sim/logger'
 import { ExternalLink, Users } from 'lucide-react'
 import { Button, Combobox } from '@/components/emcn/components'
@@ -203,7 +203,7 @@ export function CredentialSelector({
     if (!baseProviderConfig) {
       return <ExternalLink className='h-3 w-3' />
     }
-    return baseProviderConfig.icon({ className: 'h-3 w-3' })
+    return createElement(baseProviderConfig.icon, { className: 'h-3 w-3' })
   }, [])
 
   const getProviderName = useCallback((providerName: OAuthProvider) => {

apps/sim/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/messages-input/messages-input.tsx

@@ -7,13 +7,24 @@ import {
   useRef,
   useState,
 } from 'react'
+import { createLogger } from '@sim/logger'
 import { isEqual } from 'lodash'
-import { ChevronDown, ChevronsUpDown, ChevronUp, Plus } from 'lucide-react'
-import { Button, Popover, PopoverContent, PopoverItem, PopoverTrigger } from '@/components/emcn'
+import { ArrowLeftRight, ChevronDown, ChevronsUpDown, ChevronUp, Plus } from 'lucide-react'
+import { useParams } from 'next/navigation'
+import {
+  Button,
+  Popover,
+  PopoverContent,
+  PopoverItem,
+  PopoverTrigger,
+  Tooltip,
+} from '@/components/emcn'
 import { Trash } from '@/components/emcn/icons/trash'
 import { cn } from '@/lib/core/utils/cn'
 import { EnvVarDropdown } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/env-var-dropdown'
+import { FileUpload } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/file-upload/file-upload'
 import { formatDisplayText } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/formatted-text'
+import { ShortInput } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/short-input/short-input'
 import { TagDropdown } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/tag-dropdown/tag-dropdown'
 import { useSubBlockInput } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/hooks/use-sub-block-input'
 import { useSubBlockValue } from '@/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/hooks/use-sub-block-value'
@@ -21,19 +32,32 @@ import type { WandControlHandlers } from '@/app/workspace/[workspaceId]/w/[workf
 import { useAccessibleReferencePrefixes } from '@/app/workspace/[workspaceId]/w/[workflowId]/hooks/use-accessible-reference-prefixes'
 import { useWand } from '@/app/workspace/[workspaceId]/w/[workflowId]/hooks/use-wand'
 import type { SubBlockConfig } from '@/blocks/types'
+import { supportsVision } from '@/providers/utils'
+import { useWorkflowRegistry } from '@/stores/workflows/registry/store'
+import { useSubBlockStore } from '@/stores/workflows/subblock/store'
+
+const logger = createLogger('MessagesInput')
 
 const MIN_TEXTAREA_HEIGHT_PX = 80
+
+/** Workspace file record from API */
+interface WorkspaceFile {
+  id: string
+  name: string
+  path: string
+  type: string
+}
 const MAX_TEXTAREA_HEIGHT_PX = 320
 
 /** Pattern to match complete message objects in JSON */
 const COMPLETE_MESSAGE_PATTERN =
-  /"role"\s*:\s*"(system|user|assistant)"[^}]*"content"\s*:\s*"((?:[^"\\]|\\.)*)"/g
+  /"role"\s*:\s*"(system|user|assistant|attachment)"[^}]*"content"\s*:\s*"((?:[^"\\]|\\.)*)"/g
 
 /** Pattern to match incomplete content at end of buffer */
 const INCOMPLETE_CONTENT_PATTERN = /"content"\s*:\s*"((?:[^"\\]|\\.)*)$/
 
 /** Pattern to match role before content */
-const ROLE_BEFORE_CONTENT_PATTERN = /"role"\s*:\s*"(system|user|assistant)"[^{]*$/
+const ROLE_BEFORE_CONTENT_PATTERN = /"role"\s*:\s*"(system|user|assistant|attachment)"[^{]*$/
 
 /**
  * Unescapes JSON string content
@@ -41,41 +65,46 @@ const ROLE_BEFORE_CONTENT_PATTERN = /"role"\s*:\s*"(system|user|assistant)"[^{]*
 const unescapeContent = (str: string): string =>
   str.replace(/\\n/g, '\n').replace(/\\"/g, '"').replace(/\\\\/g, '\\')
 
+/**
+ * Attachment content (files, images, documents)
+ */
+interface AttachmentContent {
+  /** Source type: how the data was provided */
+  sourceType: 'url' | 'base64' | 'file'
+  /** The URL or base64 data */
+  data: string
+  /** MIME type (e.g., 'image/png', 'application/pdf', 'audio/mp3') */
+  mimeType?: string
+  /** Optional filename for file uploads */
+  fileName?: string
+  /** Optional workspace file ID (used by wand to select existing files) */
+  fileId?: string
+}
+
 /**
  * Interface for individual message in the messages array
  */
 interface Message {
-  role: 'system' | 'user' | 'assistant'
+  role: 'system' | 'user' | 'assistant' | 'attachment'
   content: string
+  attachment?: AttachmentContent
 }
 
 /**
  * Props for the MessagesInput component
  */
 interface MessagesInputProps {
-  /** Unique identifier for the block */
   blockId: string
-  /** Unique identifier for the sub-block */
   subBlockId: string
-  /** Configuration object for the sub-block */
   config: SubBlockConfig
-  /** Whether component is in preview mode */
   isPreview?: boolean
-  /** Value to display in preview mode */
   previewValue?: Message[] | null
-  /** Whether the input is disabled */
   disabled?: boolean
-  /** Ref to expose wand control handlers to parent */
   wandControlRef?: React.MutableRefObject<WandControlHandlers | null>
 }
 
 /**
  * MessagesInput component for managing LLM message history
- *
- * @remarks
- * - Manages an array of messages with role and content
- * - Each message can be edited, removed, or reordered
- * - Stores data in LLM-compatible format: [{ role, content }]
  */
 export function MessagesInput({
   blockId,
@@ -86,10 +115,163 @@ export function MessagesInput({
   disabled = false,
   wandControlRef,
 }: MessagesInputProps) {
+  const params = useParams()
+  const workspaceId = params?.workspaceId as string
   const [messages, setMessages] = useSubBlockValue<Message[]>(blockId, subBlockId, false)
   const [localMessages, setLocalMessages] = useState<Message[]>([{ role: 'user', content: '' }])
   const accessiblePrefixes = useAccessibleReferencePrefixes(blockId)
   const [openPopoverIndex, setOpenPopoverIndex] = useState<number | null>(null)
+  const { activeWorkflowId } = useWorkflowRegistry()
+
+  // Local attachment mode state - basic = FileUpload, advanced = URL/base64 textarea
+  const [attachmentMode, setAttachmentMode] = useState<'basic' | 'advanced'>('basic')
+
+  // Workspace files for wand context
+  const [workspaceFiles, setWorkspaceFiles] = useState<WorkspaceFile[]>([])
+
+  // Fetch workspace files for wand context
+  const loadWorkspaceFiles = useCallback(async () => {
+    if (!workspaceId || isPreview) return
+
+    try {
+      const response = await fetch(`/api/workspaces/${workspaceId}/files`)
+      const data = await response.json()
+      if (data.success) {
+        setWorkspaceFiles(data.files || [])
+      }
+    } catch (error) {
+      logger.error('Error loading workspace files:', error)
+    }
+  }, [workspaceId, isPreview])
+
+  // Load workspace files on mount
+  useEffect(() => {
+    void loadWorkspaceFiles()
+  }, [loadWorkspaceFiles])
+
+  // Build sources string for wand - available workspace files
+  const sourcesInfo = useMemo(() => {
+    if (workspaceFiles.length === 0) {
+      return 'No workspace files available. The user can upload files manually after generation.'
+    }
+
+    const filesList = workspaceFiles
+      .filter(
+        (f) =>
+          f.type.startsWith('image/') ||
+          f.type.startsWith('audio/') ||
+          f.type.startsWith('video/') ||
+          f.type === 'application/pdf'
+      )
+      .map((f) => `  - id: "${f.id}", name: "${f.name}", type: "${f.type}"`)
+      .join('\n')
+
+    if (!filesList) {
+      return 'No files in workspace. The user can upload files manually after generation.'
+    }
+
+    return `AVAILABLE WORKSPACE FILES (optional - you don't have to select one):\n${filesList}\n\nTo use a file, include "fileId": "<id>" in the attachment object. If not selecting a file, omit the fileId field.`
+  }, [workspaceFiles])
+
+  // Get indices of attachment messages for subscription
+  const attachmentIndices = useMemo(
+    () =>
+      localMessages
+        .map((msg, index) => (msg.role === 'attachment' ? index : -1))
+        .filter((i) => i !== -1),
+    [localMessages]
+  )
+
+  // Subscribe to model value to check vision capability
+  const modelSupportsVision = useSubBlockStore(
+    useCallback(
+      (state) => {
+        if (!activeWorkflowId) return true // Default to allowing attachments
+        const blockValues = state.workflowValues[activeWorkflowId]?.[blockId] ?? {}
+        const modelValue = blockValues.model as string | undefined
+        if (!modelValue) return true // No model selected, allow attachments
+        return supportsVision(modelValue)
+      },
+      [activeWorkflowId, blockId]
+    )
+  )
+
+  // Determine available roles based on model capabilities
+  const availableRoles = useMemo(() => {
+    const baseRoles: Array<'system' | 'user' | 'assistant' | 'attachment'> = [
+      'system',
+      'user',
+      'assistant',
+    ]
+    if (modelSupportsVision) {
+      baseRoles.push('attachment')
+    }
+    return baseRoles
+  }, [modelSupportsVision])
+
+  // Subscribe to file upload values for all attachment messages
+  const fileUploadValues = useSubBlockStore(
+    useCallback(
+      (state) => {
+        if (!activeWorkflowId) return {}
+        const blockValues = state.workflowValues[activeWorkflowId]?.[blockId] ?? {}
+        const result: Record<number, { name: string; path: string; type: string; size: number }> =
+          {}
+        for (const index of attachmentIndices) {
+          const fileUploadKey = `${subBlockId}-attachment-${index}`
+          const fileValue = blockValues[fileUploadKey]
+          if (fileValue && typeof fileValue === 'object' && 'path' in fileValue) {
+            result[index] = fileValue as { name: string; path: string; type: string; size: number }
+          }
+        }
+        return result
+      },
+      [activeWorkflowId, blockId, subBlockId, attachmentIndices]
+    )
+  )
+
+  // Effect to sync FileUpload values to message attachment objects
+  useEffect(() => {
+    if (!activeWorkflowId || isPreview) return
+
+    let hasChanges = false
+    const updatedMessages = localMessages.map((msg, index) => {
+      if (msg.role !== 'attachment') return msg
+
+      const uploadedFile = fileUploadValues[index]
+      if (uploadedFile) {
+        const newAttachment: AttachmentContent = {
+          sourceType: 'file',
+          data: uploadedFile.path,
+          mimeType: uploadedFile.type,
+          fileName: uploadedFile.name,
+        }
+
+        // Only update if different
+        if (
+          msg.attachment?.data !== newAttachment.data ||
+          msg.attachment?.sourceType !== newAttachment.sourceType ||
+          msg.attachment?.mimeType !== newAttachment.mimeType ||
+          msg.attachment?.fileName !== newAttachment.fileName
+        ) {
+          hasChanges = true
+          return {
+            ...msg,
+            content: uploadedFile.name || msg.content,
+            attachment: newAttachment,
+          }
+        }
+      }
+
+      return msg
+    })
+
+    if (hasChanges) {
+      setLocalMessages(updatedMessages)
+      setMessages(updatedMessages)
+    }
+  }, [activeWorkflowId, localMessages, isPreview, setMessages, fileUploadValues])
+
   const subBlockInput = useSubBlockInput({
     blockId,
     subBlockId,
@@ -98,43 +280,40 @@ export function MessagesInput({
     disabled,
   })
 
-  /**
-   * Gets the current messages as JSON string for wand context
-   */
   const getMessagesJson = useCallback((): string => {
     if (localMessages.length === 0) return ''
-    // Filter out empty messages for cleaner context
     const nonEmptyMessages = localMessages.filter((m) => m.content.trim() !== '')
     if (nonEmptyMessages.length === 0) return ''
     return JSON.stringify(nonEmptyMessages, null, 2)
   }, [localMessages])
 
-  /**
-   * Streaming buffer for accumulating JSON content
-   */
   const streamBufferRef = useRef<string>('')
 
-  /**
-   * Parses and validates messages from JSON content
-   */
   const parseMessages = useCallback((content: string): Message[] | null => {
     try {
       const parsed = JSON.parse(content)
       if (Array.isArray(parsed)) {
         const validMessages: Message[] = parsed
           .filter(
-            (m): m is { role: string; content: string } =>
+            (m): m is { role: string; content: string; attachment?: AttachmentContent } =>
               typeof m === 'object' &&
               m !== null &&
               typeof m.role === 'string' &&
               typeof m.content === 'string'
           )
-          .map((m) => ({
-            role: (['system', 'user', 'assistant'].includes(m.role)
+          .map((m) => {
+            const role = ['system', 'user', 'assistant', 'attachment'].includes(m.role)
               ? m.role
-              : 'user') as Message['role'],
-            content: m.content,
-          }))
+              : 'user'
+            const message: Message = {
+              role: role as Message['role'],
+              content: m.content,
+            }
+            if (m.attachment) {
+              message.attachment = m.attachment
+            }
+            return message
+          })
         return validMessages.length > 0 ? validMessages : null
       }
     } catch {
@@ -143,26 +322,19 @@ export function MessagesInput({
     return null
   }, [])
 
-  /**
-   * Extracts messages from streaming JSON buffer
-   * Uses simple pattern matching for efficiency
-   */
   const extractStreamingMessages = useCallback(
     (buffer: string): Message[] => {
-      // Try complete JSON parse first
       const complete = parseMessages(buffer)
       if (complete) return complete
 
       const result: Message[] = []
 
-      // Reset regex lastIndex for global pattern
       COMPLETE_MESSAGE_PATTERN.lastIndex = 0
       let match
       while ((match = COMPLETE_MESSAGE_PATTERN.exec(buffer)) !== null) {
         result.push({ role: match[1] as Message['role'], content: unescapeContent(match[2]) })
       }
 
-      // Check for incomplete message at end (content still streaming)
       const lastContentIdx = buffer.lastIndexOf('"content"')
       if (lastContentIdx !== -1) {
         const tail = buffer.slice(lastContentIdx)
@@ -172,7 +344,6 @@ export function MessagesInput({
           const roleMatch = head.match(ROLE_BEFORE_CONTENT_PATTERN)
           if (roleMatch) {
             const content = unescapeContent(incomplete[1])
-            // Only add if not duplicate of last complete message
             if (result.length === 0 || result[result.length - 1].content !== content) {
               result.push({ role: roleMatch[1] as Message['role'], content })
             }
@@ -185,12 +356,10 @@ export function MessagesInput({
     [parseMessages]
   )
 
-  /**
-   * Wand hook for AI-assisted content generation
-   */
   const wandHook = useWand({
     wandConfig: config.wandConfig,
     currentValue: getMessagesJson(),
+    sources: sourcesInfo,
     onStreamStart: () => {
       streamBufferRef.current = ''
       setLocalMessages([{ role: 'system', content: '' }])
@@ -205,10 +374,50 @@ export function MessagesInput({
     onGeneratedContent: (content) => {
       const validMessages = parseMessages(content)
       if (validMessages) {
+        // Process attachment messages - only allow fileId to set files, sanitize other attempts
+        validMessages.forEach((msg, index) => {
+          if (msg.role === 'attachment') {
+            // Check if this is an existing file with valid data (preserve it)
+            const hasExistingFile =
+              msg.attachment?.sourceType === 'file' &&
+              msg.attachment?.data?.startsWith('/api/') &&
+              msg.attachment?.fileName
+
+            if (hasExistingFile) {
+              // Preserve existing file data as-is
+              return
+            }
+
+            // Check if wand provided a fileId to select a workspace file
+            if (msg.attachment?.fileId) {
+              const file = workspaceFiles.find((f) => f.id === msg.attachment?.fileId)
+              if (file) {
+                // Set the file value in SubBlockStore so FileUpload picks it up
+                const fileUploadKey = `${subBlockId}-attachment-${index}`
+                const uploadedFile = {
+                  name: file.name,
+                  path: file.path,
+                  type: file.type,
+                  size: 0, // Size not available from workspace files list
+                }
+                useSubBlockStore.getState().setValue(blockId, fileUploadKey, uploadedFile)
+
+                // Clear the attachment object - the FileUpload will sync the file data via useEffect
+                // DON'T set attachment.data here as it would appear in the ShortInput (advanced mode)
+                msg.attachment = undefined
+                return
+              }
+            }
+
+            // Sanitize: clear any attachment object that isn't a valid existing file or fileId match
+            // This prevents the LLM from setting arbitrary data/variable references
+            msg.attachment = undefined
+          }
+        })
+
         setLocalMessages(validMessages)
         setMessages(validMessages)
       } else {
-        // Fallback: treat as raw system prompt
         const trimmed = content.trim()
         if (trimmed) {
           const fallback: Message[] = [{ role: 'system', content: trimmed }]
@@ -219,9 +428,6 @@ export function MessagesInput({
     },
   })
 
-  /**
-   * Expose wand control handlers to parent via ref
-   */
   useImperativeHandle(
     wandControlRef,
     () => ({
@@ -249,9 +455,6 @@ export function MessagesInput({
     }
   }, [isPreview, previewValue, messages])
 
-  /**
-   * Gets the current messages array
-   */
   const currentMessages = useMemo<Message[]>(() => {
     if (isPreview && previewValue && Array.isArray(previewValue)) {
       return previewValue
@@ -269,9 +472,6 @@ export function MessagesInput({
     startHeight: number
   } | null>(null)
 
-  /**
-   * Updates a specific message's content
-   */
   const updateMessageContent = useCallback(
     (index: number, content: string) => {
       if (isPreview || disabled) return
@@ -287,17 +487,27 @@ export function MessagesInput({
     [localMessages, setMessages, isPreview, disabled]
   )
 
-  /**
-   * Updates a specific message's role
-   */
   const updateMessageRole = useCallback(
-    (index: number, role: 'system' | 'user' | 'assistant') => {
+    (index: number, role: 'system' | 'user' | 'assistant' | 'attachment') => {
       if (isPreview || disabled) return
 
       const updatedMessages = [...localMessages]
-      updatedMessages[index] = {
-        ...updatedMessages[index],
-        role,
+      if (role === 'attachment') {
+        updatedMessages[index] = {
+          ...updatedMessages[index],
+          role,
+          content: updatedMessages[index].content || '',
+          attachment: updatedMessages[index].attachment || {
+            sourceType: 'file',
+            data: '',
+          },
+        }
+      } else {
+        const { attachment: _, ...rest } = updatedMessages[index]
+        updatedMessages[index] = {
+          ...rest,
+          role,
+        }
       }
       setLocalMessages(updatedMessages)
       setMessages(updatedMessages)
@@ -305,9 +515,6 @@ export function MessagesInput({
     [localMessages, setMessages, isPreview, disabled]
   )
 
-  /**
-   * Adds a message after the specified index
-   */
   const addMessageAfter = useCallback(
     (index: number) => {
       if (isPreview || disabled) return
@@ -320,9 +527,6 @@ export function MessagesInput({
     [localMessages, setMessages, isPreview, disabled]
   )
 
-  /**
-   * Deletes a message at the specified index
-   */
   const deleteMessage = useCallback(
     (index: number) => {
       if (isPreview || disabled) return
@@ -335,9 +539,6 @@ export function MessagesInput({
     [localMessages, setMessages, isPreview, disabled]
   )
 
-  /**
-   * Moves a message up in the list
-   */
   const moveMessageUp = useCallback(
     (index: number) => {
       if (isPreview || disabled || index === 0) return
@@ -352,9 +553,6 @@ export function MessagesInput({
     [localMessages, setMessages, isPreview, disabled]
   )
 
-  /**
-   * Moves a message down in the list
-   */
   const moveMessageDown = useCallback(
     (index: number) => {
       if (isPreview || disabled || index === localMessages.length - 1) return
@@ -369,18 +567,11 @@ export function MessagesInput({
     [localMessages, setMessages, isPreview, disabled]
   )
 
-  /**
-   * Capitalizes the first letter of the role
-   */
   const formatRole = (role: string): string => {
     return role.charAt(0).toUpperCase() + role.slice(1)
   }
 
-  /**
-   * Handles header click to focus the textarea
-   */
   const handleHeaderClick = useCallback((index: number, e: React.MouseEvent) => {
-    // Don't focus if clicking on interactive elements
     const target = e.target as HTMLElement
     if (target.closest('button') || target.closest('[data-radix-popper-content-wrapper]')) {
       return
@@ -570,50 +761,52 @@ export function MessagesInput({
                   className='flex cursor-pointer items-center justify-between px-[8px] pt-[6px]'
                   onClick={(e) => handleHeaderClick(index, e)}
                 >
-                  <Popover
-                    open={openPopoverIndex === index}
-                    onOpenChange={(open) => setOpenPopoverIndex(open ? index : null)}
-                  >
-                    <PopoverTrigger asChild>
-                      <button
-                        type='button'
-                        disabled={isPreview || disabled}
-                        className={cn(
-                          'group -ml-1.5 -my-1 flex items-center gap-1 rounded px-1.5 py-1 font-medium text-[13px] text-[var(--text-primary)] leading-none transition-colors hover:bg-[var(--surface-5)] hover:text-[var(--text-secondary)]',
-                          (isPreview || disabled) &&
-                            'cursor-default hover:bg-transparent hover:text-[var(--text-primary)]'
-                        )}
-                        onClick={(e) => e.stopPropagation()}
-                        aria-label='Select message role'
-                      >
-                        {formatRole(message.role)}
-                        {!isPreview && !disabled && (
-                          <ChevronDown
-                            className={cn(
-                              'h-3 w-3 flex-shrink-0 transition-transform duration-100',
-                              openPopoverIndex === index && 'rotate-180'
-                            )}
-                          />
-                        )}
-                      </button>
-                    </PopoverTrigger>
-                    <PopoverContent minWidth={140} align='start'>
-                      <div className='flex flex-col gap-[2px]'>
-                        {(['system', 'user', 'assistant'] as const).map((role) => (
-                          <PopoverItem
-                            key={role}
-                            active={message.role === role}
-                            onClick={() => {
-                              updateMessageRole(index, role)
-                              setOpenPopoverIndex(null)
-                            }}
-                          >
-                            <span>{formatRole(role)}</span>
-                          </PopoverItem>
-                        ))}
-                      </div>
-                    </PopoverContent>
-                  </Popover>
+                  <div className='flex items-center'>
+                    <Popover
+                      open={openPopoverIndex === index}
+                      onOpenChange={(open) => setOpenPopoverIndex(open ? index : null)}
+                    >
+                      <PopoverTrigger asChild>
+                        <button
+                          type='button'
+                          disabled={isPreview || disabled}
+                          className={cn(
+                            'group -ml-1.5 -my-1 flex items-center gap-1 rounded px-1.5 py-1 font-medium text-[13px] text-[var(--text-primary)] leading-none transition-colors hover:bg-[var(--surface-5)] hover:text-[var(--text-secondary)]',
+                            (isPreview || disabled) &&
+                              'cursor-default hover:bg-transparent hover:text-[var(--text-primary)]'
+                          )}
+                          onClick={(e) => e.stopPropagation()}
+                          aria-label='Select message role'
+                        >
+                          {formatRole(message.role)}
+                          {!isPreview && !disabled && (
+                            <ChevronDown
+                              className={cn(
+                                'h-3 w-3 flex-shrink-0 transition-transform duration-100',
+                                openPopoverIndex === index && 'rotate-180'
+                              )}
+                            />
+                          )}
+                        </button>
+                      </PopoverTrigger>
+                      <PopoverContent minWidth={140} align='start'>
+                        <div className='flex flex-col gap-[2px]'>
+                          {availableRoles.map((role) => (
+                            <PopoverItem
+                              key={role}
+                              active={message.role === role}
+                              onClick={() => {
+                                updateMessageRole(index, role)
+                                setOpenPopoverIndex(null)
+                              }}
+                            >
+                              <span>{formatRole(role)}</span>
+                            </PopoverItem>
+                          ))}
+                        </div>
+                      </PopoverContent>
+                    </Popover>
+                  </div>
 
                   {!isPreview && !disabled && (
                     <div className='flex items-center'>
@@ -657,6 +850,43 @@ export function MessagesInput({
                           </Button>
                         </>
                       )}
+                      {/* Mode toggle for attachment messages */}
+                      {message.role === 'attachment' && (
+                        <Tooltip.Root>
+                          <Tooltip.Trigger asChild>
+                            <Button
+                              variant='ghost'
+                              onClick={(e: React.MouseEvent) => {
+                                e.stopPropagation()
+                                setAttachmentMode((m) => (m === 'basic' ? 'advanced' : 'basic'))
+                              }}
+                              disabled={disabled}
+                              className='-my-1 -mr-1 h-6 w-6 p-0'
+                              aria-label={
+                                attachmentMode === 'advanced'
+                                  ? 'Switch to file upload'
+                                  : 'Switch to URL/text input'
+                              }
+                            >
+                              <ArrowLeftRight
+                                className={cn(
+                                  'h-3 w-3',
+                                  attachmentMode === 'advanced'
+                                    ? 'text-[var(--text-primary)]'
+                                    : 'text-[var(--text-secondary)]'
+                                )}
+                              />
+                            </Button>
+                          </Tooltip.Trigger>
+                          <Tooltip.Content side='top'>
+                            <p>
+                              {attachmentMode === 'advanced'
+                                ? 'Switch to file upload'
+                                : 'Switch to URL/text input'}
+                            </p>
+                          </Tooltip.Content>
+                        </Tooltip.Root>
+                      )}
                       <Button
                         variant='ghost'
                         onClick={(e: React.MouseEvent) => {
@@ -673,98 +903,152 @@ export function MessagesInput({
                   )}
                 </div>
 
-                {/* Content Input with overlay for variable highlighting */}
-                <div className='relative w-full overflow-hidden'>
-                  <textarea
-                    ref={(el) => {
-                      textareaRefs.current[fieldId] = el
-                    }}
-                    className='relative z-[2] m-0 box-border h-auto min-h-[80px] w-full resize-none overflow-y-auto overflow-x-hidden whitespace-pre-wrap break-words border-none bg-transparent px-[8px] py-[8px] font-medium font-sans text-sm text-transparent leading-[1.5] caret-[var(--text-primary)] outline-none [-ms-overflow-style:none] [scrollbar-width:none] placeholder:text-[var(--text-muted)] focus:outline-none focus-visible:outline-none disabled:cursor-not-allowed [&::-webkit-scrollbar]:hidden'
-                    placeholder='Enter message content...'
-                    value={message.content}
-                    onChange={fieldHandlers.onChange}
-                    onKeyDown={(e) => {
-                      if (e.key === 'Tab' && !isPreview && !disabled) {
-                        e.preventDefault()
-                        const direction = e.shiftKey ? -1 : 1
-                        const nextIndex = index + direction
-
-                        if (nextIndex >= 0 && nextIndex < currentMessages.length) {
-                          const nextFieldId = `message-${nextIndex}`
-                          const nextTextarea = textareaRefs.current[nextFieldId]
-                          if (nextTextarea) {
-                            nextTextarea.focus()
-                            nextTextarea.selectionStart = nextTextarea.value.length
-                            nextTextarea.selectionEnd = nextTextarea.value.length
-                          }
+                {/* Content Input - different for attachment vs text messages */}
+                {message.role === 'attachment' ? (
+                  <div className='relative w-full px-[8px] py-[8px]'>
+                    {attachmentMode === 'basic' ? (
+                      <FileUpload
+                        blockId={blockId}
+                        subBlockId={`${subBlockId}-attachment-${index}`}
+                        acceptedTypes='image/*,audio/*,video/*,application/pdf,.doc,.docx,.txt'
+                        multiple={false}
+                        isPreview={isPreview}
+                        disabled={disabled}
+                      />
+                    ) : (
+                      <ShortInput
+                        blockId={blockId}
+                        subBlockId={`${subBlockId}-attachment-ref-${index}`}
+                        placeholder='Reference file from previous block...'
+                        config={{
+                          id: `${subBlockId}-attachment-ref-${index}`,
+                          type: 'short-input',
+                        }}
+                        value={
+                          // Only show value for variable references, not file uploads
+                          message.attachment?.sourceType === 'file'
+                            ? ''
+                            : message.attachment?.data || ''
                         }
-                        return
-                      }
-
-                      fieldHandlers.onKeyDown(e)
-                    }}
-                    onDrop={fieldHandlers.onDrop}
-                    onDragOver={fieldHandlers.onDragOver}
-                    onFocus={fieldHandlers.onFocus}
-                    onScroll={(e) => {
-                      const overlay = overlayRefs.current[fieldId]
-                      if (overlay) {
-                        overlay.scrollTop = e.currentTarget.scrollTop
-                        overlay.scrollLeft = e.currentTarget.scrollLeft
-                      }
-                    }}
-                    disabled={isPreview || disabled}
-                  />
-                  <div
-                    ref={(el) => {
-                      overlayRefs.current[fieldId] = el
-                    }}
-                    className='pointer-events-none absolute top-0 left-0 z-[1] m-0 box-border w-full overflow-y-auto overflow-x-hidden whitespace-pre-wrap break-words border-none bg-transparent px-[8px] py-[8px] font-medium font-sans text-[var(--text-primary)] text-sm leading-[1.5] [-ms-overflow-style:none] [scrollbar-width:none] [&::-webkit-scrollbar]:hidden'
-                  >
-                    {formatDisplayText(message.content, {
-                      accessiblePrefixes,
-                      highlightAll: !accessiblePrefixes,
-                    })}
-                    {message.content.endsWith('\n') && '\u200B'}
+                        onChange={(newValue: string) => {
+                          const updatedMessages = [...localMessages]
+                          if (updatedMessages[index].role === 'attachment') {
+                            // Determine sourceType based on content
+                            let sourceType: 'url' | 'base64' = 'url'
+                            if (newValue.startsWith('data:') || newValue.includes(';base64,')) {
+                              sourceType = 'base64'
+                            }
+                            updatedMessages[index] = {
+                              ...updatedMessages[index],
+                              content: newValue.substring(0, 50),
+                              attachment: {
+                                ...updatedMessages[index].attachment,
+                                sourceType,
+                                data: newValue,
+                              },
+                            }
+                            setLocalMessages(updatedMessages)
+                            setMessages(updatedMessages)
+                          }
+                        }}
+                        isPreview={isPreview}
+                        disabled={disabled}
+                      />
+                    )}
                   </div>
-
-                  {/* Env var dropdown for this message */}
-                  <EnvVarDropdown
-                    visible={fieldState.showEnvVars && !isPreview && !disabled}
-                    onSelect={handleEnvSelect}
-                    searchTerm={fieldState.searchTerm}
-                    inputValue={message.content}
-                    cursorPosition={fieldState.cursorPosition}
-                    onClose={() => subBlockInput.fieldHelpers.hideFieldDropdowns(fieldId)}
-                    workspaceId={subBlockInput.workspaceId}
-                    maxHeight='192px'
-                    inputRef={textareaRefObject}
-                  />
-
-                  {/* Tag dropdown for this message */}
-                  <TagDropdown
-                    visible={fieldState.showTags && !isPreview && !disabled}
-                    onSelect={handleTagSelect}
-                    blockId={blockId}
-                    activeSourceBlockId={fieldState.activeSourceBlockId}
-                    inputValue={message.content}
-                    cursorPosition={fieldState.cursorPosition}
-                    onClose={() => subBlockInput.fieldHelpers.hideFieldDropdowns(fieldId)}
-                    inputRef={textareaRefObject}
-                  />
-
-                  {!isPreview && !disabled && (
-                    <div
-                      className='absolute right-1 bottom-1 z-[3] flex h-4 w-4 cursor-ns-resize items-center justify-center rounded-[4px] border border-[var(--border-1)] bg-[var(--surface-5)] dark:bg-[var(--surface-5)]'
-                      onMouseDown={(e) => handleResizeStart(fieldId, e)}
-                      onDragStart={(e) => {
-                        e.preventDefault()
+                ) : (
+                  <div className='relative w-full overflow-hidden'>
+                    <textarea
+                      ref={(el) => {
+                        textareaRefs.current[fieldId] = el
                       }}
+                      className='relative z-[2] m-0 box-border h-auto min-h-[80px] w-full resize-none overflow-y-auto overflow-x-hidden whitespace-pre-wrap break-words border-none bg-transparent px-[8px] py-[8px] font-medium font-sans text-sm text-transparent leading-[1.5] caret-[var(--text-primary)] outline-none [-ms-overflow-style:none] [scrollbar-width:none] placeholder:text-[var(--text-muted)] focus:outline-none focus-visible:outline-none disabled:cursor-not-allowed [&::-webkit-scrollbar]:hidden'
+                      placeholder='Enter message content...'
+                      value={message.content}
+                      onChange={fieldHandlers.onChange}
+                      onKeyDown={(e) => {
+                        if (e.key === 'Tab' && !isPreview && !disabled) {
+                          e.preventDefault()
+                          const direction = e.shiftKey ? -1 : 1
+                          const nextIndex = index + direction
+
+                          if (nextIndex >= 0 && nextIndex < currentMessages.length) {
+                            const nextFieldId = `message-${nextIndex}`
+                            const nextTextarea = textareaRefs.current[nextFieldId]
+                            if (nextTextarea) {
+                              nextTextarea.focus()
+                              nextTextarea.selectionStart = nextTextarea.value.length
+                              nextTextarea.selectionEnd = nextTextarea.value.length
+                            }
+                          }
+                          return
+                        }
+
+                        fieldHandlers.onKeyDown(e)
+                      }}
+                      onDrop={fieldHandlers.onDrop}
+                      onDragOver={fieldHandlers.onDragOver}
+                      onFocus={fieldHandlers.onFocus}
+                      onScroll={(e) => {
+                        const overlay = overlayRefs.current[fieldId]
+                        if (overlay) {
+                          overlay.scrollTop = e.currentTarget.scrollTop
+                          overlay.scrollLeft = e.currentTarget.scrollLeft
+                        }
+                      }}
+                      disabled={isPreview || disabled}
+                    />
+                    <div
+                      ref={(el) => {
+                        overlayRefs.current[fieldId] = el
+                      }}
+                      className='pointer-events-none absolute top-0 left-0 z-[1] m-0 box-border w-full overflow-y-auto overflow-x-hidden whitespace-pre-wrap break-words border-none bg-transparent px-[8px] py-[8px] font-medium font-sans text-[var(--text-primary)] text-sm leading-[1.5] [-ms-overflow-style:none] [scrollbar-width:none] [&::-webkit-scrollbar]:hidden'
                     >
-                      <ChevronsUpDown className='h-3 w-3 text-[var(--text-muted)]' />
+                      {formatDisplayText(message.content, {
+                        accessiblePrefixes,
+                        highlightAll: !accessiblePrefixes,
+                      })}
+                      {message.content.endsWith('\n') && '\u200B'}
                     </div>
-                  )}
-                </div>
+
+                    {/* Env var dropdown for this message */}
+                    <EnvVarDropdown
+                      visible={fieldState.showEnvVars && !isPreview && !disabled}
+                      onSelect={handleEnvSelect}
+                      searchTerm={fieldState.searchTerm}
+                      inputValue={message.content}
+                      cursorPosition={fieldState.cursorPosition}
+                      onClose={() => subBlockInput.fieldHelpers.hideFieldDropdowns(fieldId)}
+                      workspaceId={subBlockInput.workspaceId}
+                      maxHeight='192px'
+                      inputRef={textareaRefObject}
+                    />
+
+                    {/* Tag dropdown for this message */}
+                    <TagDropdown
+                      visible={fieldState.showTags && !isPreview && !disabled}
+                      onSelect={handleTagSelect}
+                      blockId={blockId}
+                      activeSourceBlockId={fieldState.activeSourceBlockId}
+                      inputValue={message.content}
+                      cursorPosition={fieldState.cursorPosition}
+                      onClose={() => subBlockInput.fieldHelpers.hideFieldDropdowns(fieldId)}
+                      inputRef={textareaRefObject}
+                    />
+
+                    {!isPreview && !disabled && (
+                      <div
+                        className='absolute right-1 bottom-1 z-[3] flex h-4 w-4 cursor-ns-resize items-center justify-center rounded-[4px] border border-[var(--border-1)] bg-[var(--surface-5)] dark:bg-[var(--surface-5)]'
+                        onMouseDown={(e) => handleResizeStart(fieldId, e)}
+                        onDragStart={(e) => {
+                          e.preventDefault()
+                        }}
+                      >
+                        <ChevronsUpDown className='h-3 w-3 text-[var(--text-muted)]' />
+                      </div>
+                    )}
+                  </div>
+                )}
               </>
             )
           })()}

apps/sim/app/workspace/[workspaceId]/w/[workflowId]/components/panel/components/editor/components/sub-block/components/tool-input/components/tool-credential-selector.tsx

@@ -1,4 +1,4 @@
-import { useCallback, useEffect, useMemo, useState } from 'react'
+import { createElement, useCallback, useEffect, useMemo, useState } from 'react'
 import { ExternalLink } from 'lucide-react'
 import { Button, Combobox } from '@/components/emcn/components'
 import {
@@ -22,7 +22,7 @@ const getProviderIcon = (providerName: OAuthProvider) => {
   if (!baseProviderConfig) {
     return <ExternalLink className='h-3 w-3' />
   }
-  return baseProviderConfig.icon({ className: 'h-3 w-3' })
+  return createElement(baseProviderConfig.icon, { className: 'h-3 w-3' })
 }
 
 const getProviderName = (providerName: OAuthProvider) => {

apps/sim/app/workspace/[workspaceId]/w/[workflowId]/hooks/use-wand.ts

@@ -63,6 +63,8 @@ export interface WandConfig {
 interface UseWandProps {
   wandConfig?: WandConfig
   currentValue?: string
+  /** Additional context about available sources/references for the prompt */
+  sources?: string
   onGeneratedContent: (content: string) => void
   onStreamChunk?: (chunk: string) => void
   onStreamStart?: () => void
@@ -72,6 +74,7 @@ interface UseWandProps {
 export function useWand({
   wandConfig,
   currentValue,
+  sources,
   onGeneratedContent,
   onStreamChunk,
   onStreamStart,
@@ -154,6 +157,12 @@ export function useWand({
         if (systemPrompt.includes('{context}')) {
           systemPrompt = systemPrompt.replace('{context}', contextInfo)
         }
+        if (systemPrompt.includes('{sources}')) {
+          systemPrompt = systemPrompt.replace(
+            '{sources}',
+            sources || 'No upstream sources available'
+          )
+        }
 
         const userMessage = prompt
 

apps/sim/app/workspace/[workspaceId]/w/components/sidebar/components/settings-modal/components/integrations/integrations.tsx

@@ -1,6 +1,6 @@
 'use client'
 
-import { useEffect, useRef, useState } from 'react'
+import { createElement, useEffect, useRef, useState } from 'react'
 import { createLogger } from '@sim/logger'
 import { Check, ChevronDown, ExternalLink, Search } from 'lucide-react'
 import { useRouter, useSearchParams } from 'next/navigation'
@@ -339,9 +339,7 @@ export function Integrations({ onOpenChange, registerCloseHandler }: Integration
                     >
                       <div className='flex items-center gap-[12px]'>
                         <div className='flex h-9 w-9 flex-shrink-0 items-center justify-center overflow-hidden rounded-[6px] bg-[var(--surface-5)]'>
-                          {typeof service.icon === 'function'
-                            ? service.icon({ className: 'h-4 w-4' })
-                            : service.icon}
+                          {createElement(service.icon, { className: 'h-4 w-4' })}
                         </div>
                         <div className='flex flex-col justify-center gap-[1px]'>
                           <span className='font-medium text-[14px]'>{service.name}</span>

apps/sim/blocks/blocks/agent.ts

@@ -85,7 +85,9 @@ export const AgentBlock: BlockConfig<AgentResponse> = {
       id: 'messages',
       title: 'Messages',
       type: 'messages-input',
+      canonicalParamId: 'messages',
       placeholder: 'Enter messages...',
+      mode: 'basic',
       wandConfig: {
         enabled: true,
         maintainHistory: true,
@@ -93,10 +95,12 @@ export const AgentBlock: BlockConfig<AgentResponse> = {
 
 Current messages: {context}
 
+{sources}
+
 RULES:
 1. Generate ONLY a valid JSON array - no markdown, no explanations
-2. Each message object must have "role" (system/user/assistant) and "content" (string)
-3. You can generate any number of messages as needed
+2. Each message object must have "role" and "content" properties
+3. Valid roles are: "system", "user", "assistant", "attachment"
 4. Content can be as long as necessary - don't truncate
 5. If editing existing messages, preserve structure unless asked to change it
 6. For new agents, create DETAILED, PROFESSIONAL system prompts that include:
@@ -106,6 +110,16 @@ RULES:
    - Critical thinking or quality guidelines
    - How to handle edge cases and uncertainty
 
+ATTACHMENTS:
+- Use role "attachment" to include images, audio, video, or documents in a multimodal conversation
+- IMPORTANT: If an attachment message in the current context has an "attachment" object with file data, ALWAYS preserve that entire "attachment" object exactly as-is
+- When creating NEW attachment messages, you can either:
+  1. Just set role to "attachment" with descriptive content - user will upload the file manually
+  2. Select a file from the available workspace files by including "fileId" in the attachment object (optional)
+- You do NOT have to select a file - it's completely optional
+- Example without file: {"role": "attachment", "content": "Analyze this image for text and objects"}
+- Example with file selection: {"role": "attachment", "content": "Analyze this image", "attachment": {"fileId": "abc123"}}
+
 EXAMPLES:
 
 Research agent:
@@ -114,14 +128,23 @@ Research agent:
 Code reviewer:
 [{"role": "system", "content": "You are a Senior Code Reviewer with expertise in software architecture, security, and best practices. Your role is to provide thorough, constructive code reviews that improve code quality and help developers grow.\\n\\n## Review Methodology\\n\\n1. **Security First**: Check for vulnerabilities including injection attacks, authentication flaws, data exposure, and insecure dependencies.\\n\\n2. **Code Quality**: Evaluate readability, maintainability, adherence to DRY/SOLID principles, and appropriate abstraction levels.\\n\\n3. **Performance**: Identify potential bottlenecks, unnecessary computations, memory leaks, and optimization opportunities.\\n\\n4. **Testing**: Assess test coverage, edge case handling, and testability of the code structure.\\n\\n## Output Format\\n\\n### Summary\\nBrief overview of the code's purpose and overall assessment.\\n\\n### Critical Issues\\nSecurity vulnerabilities or bugs that must be fixed before merging.\\n\\n### Improvements\\nSuggested enhancements with clear explanations of why and how.\\n\\n### Positive Aspects\\nHighlight well-written code to reinforce good practices.\\n\\nBe specific with line references. Provide code examples for suggested changes. Balance critique with encouragement."}, {"role": "user", "content": "<start.input>"}]
 
-Writing assistant:
-[{"role": "system", "content": "You are a skilled Writing Editor and Coach. Your role is to help users improve their writing through constructive feedback, editing suggestions, and guidance on style, clarity, and structure.\\n\\n## Editing Approach\\n\\n1. **Clarity**: Ensure ideas are expressed clearly and concisely. Eliminate jargon unless appropriate for the audience.\\n\\n2. **Structure**: Evaluate logical flow, paragraph organization, and transitions between ideas.\\n\\n3. **Voice & Tone**: Maintain consistency and appropriateness for the intended audience and purpose.\\n\\n4. **Grammar & Style**: Correct errors while respecting the author's voice.\\n\\n## Output Format\\n\\n### Overall Impression\\nBrief assessment of the piece's strengths and areas for improvement.\\n\\n### Structural Feedback\\nComments on organization, flow, and logical progression.\\n\\n### Line-Level Edits\\nSpecific suggestions with explanations, not just corrections.\\n\\n### Revised Version\\nWhen appropriate, provide an edited version demonstrating improvements.\\n\\nBe encouraging while honest. Explain the reasoning behind suggestions to help the writer improve."}, {"role": "user", "content": "<start.input>"}]
+Image analysis agent:
+[{"role": "system", "content": "You are an expert image analyst. Describe images in detail, identify objects, text, and patterns. Provide structured analysis."}, {"role": "attachment", "content": "Analyze this image"}]
 
 Return ONLY the JSON array.`,
         placeholder: 'Describe what you want to create or change...',
         generationType: 'json-object',
       },
     },
+    {
+      id: 'messagesRaw',
+      title: 'Messages',
+      type: 'code',
+      canonicalParamId: 'messages',
+      placeholder: '[{"role": "system", "content": "..."}, {"role": "user", "content": "..."}]',
+      language: 'json',
+      mode: 'advanced',
+    },
     {
       id: 'model',
       title: 'Model',

apps/sim/executor/execution/edge-manager.test.ts

@@ -2417,4 +2417,177 @@ describe('EdgeManager', () => {
       expect(successReady).toContain(targetId)
     })
   })
+
+  describe('Condition with loop downstream - deactivation propagation', () => {
+    it('should deactivate nodes after loop when condition branch containing loop is deactivated', () => {
+      // Scenario: condition → (if) → sentinel_start → loopBody → sentinel_end → (loop_exit) → after_loop
+      //                     → (else) → other_branch
+      // When condition takes "else" path, the entire if-branch including nodes after the loop should be deactivated
+      const conditionId = 'condition'
+      const sentinelStartId = 'sentinel-start'
+      const loopBodyId = 'loop-body'
+      const sentinelEndId = 'sentinel-end'
+      const afterLoopId = 'after-loop'
+      const otherBranchId = 'other-branch'
+
+      const conditionNode = createMockNode(conditionId, [
+        { target: sentinelStartId, sourceHandle: 'condition-if' },
+        { target: otherBranchId, sourceHandle: 'condition-else' },
+      ])
+
+      const sentinelStartNode = createMockNode(
+        sentinelStartId,
+        [{ target: loopBodyId }],
+        [conditionId]
+      )
+
+      const loopBodyNode = createMockNode(
+        loopBodyId,
+        [{ target: sentinelEndId }],
+        [sentinelStartId]
+      )
+
+      const sentinelEndNode = createMockNode(
+        sentinelEndId,
+        [
+          { target: sentinelStartId, sourceHandle: 'loop_continue' },
+          { target: afterLoopId, sourceHandle: 'loop_exit' },
+        ],
+        [loopBodyId]
+      )
+
+      const afterLoopNode = createMockNode(afterLoopId, [], [sentinelEndId])
+      const otherBranchNode = createMockNode(otherBranchId, [], [conditionId])
+
+      const nodes = new Map<string, DAGNode>([
+        [conditionId, conditionNode],
+        [sentinelStartId, sentinelStartNode],
+        [loopBodyId, loopBodyNode],
+        [sentinelEndId, sentinelEndNode],
+        [afterLoopId, afterLoopNode],
+        [otherBranchId, otherBranchNode],
+      ])
+
+      const dag = createMockDAG(nodes)
+      const edgeManager = new EdgeManager(dag)
+
+      // Condition selects "else" branch, deactivating the "if" branch (which contains the loop)
+      const readyNodes = edgeManager.processOutgoingEdges(conditionNode, { selectedOption: 'else' })
+
+      // Only otherBranch should be ready
+      expect(readyNodes).toContain(otherBranchId)
+      expect(readyNodes).not.toContain(sentinelStartId)
+
+      // afterLoop should NOT be ready - its incoming edge from sentinel_end should be deactivated
+      expect(readyNodes).not.toContain(afterLoopId)
+
+      // Verify that countActiveIncomingEdges returns 0 for afterLoop
+      // (meaning the loop_exit edge was properly deactivated)
+      // Note: isNodeReady returns true when all edges are deactivated (no pending deps),
+      // but the node won't be in readyNodes since it wasn't reached via an active path
+      expect(edgeManager.isNodeReady(afterLoopNode)).toBe(true) // All edges deactivated = no blocking deps
+    })
+
+    it('should deactivate nodes after parallel when condition branch containing parallel is deactivated', () => {
+      // Similar scenario with parallel instead of loop
+      const conditionId = 'condition'
+      const parallelStartId = 'parallel-start'
+      const parallelBodyId = 'parallel-body'
+      const parallelEndId = 'parallel-end'
+      const afterParallelId = 'after-parallel'
+      const otherBranchId = 'other-branch'
+
+      const conditionNode = createMockNode(conditionId, [
+        { target: parallelStartId, sourceHandle: 'condition-if' },
+        { target: otherBranchId, sourceHandle: 'condition-else' },
+      ])
+
+      const parallelStartNode = createMockNode(
+        parallelStartId,
+        [{ target: parallelBodyId }],
+        [conditionId]
+      )
+
+      const parallelBodyNode = createMockNode(
+        parallelBodyId,
+        [{ target: parallelEndId }],
+        [parallelStartId]
+      )
+
+      const parallelEndNode = createMockNode(
+        parallelEndId,
+        [{ target: afterParallelId, sourceHandle: 'parallel_exit' }],
+        [parallelBodyId]
+      )
+
+      const afterParallelNode = createMockNode(afterParallelId, [], [parallelEndId])
+      const otherBranchNode = createMockNode(otherBranchId, [], [conditionId])
+
+      const nodes = new Map<string, DAGNode>([
+        [conditionId, conditionNode],
+        [parallelStartId, parallelStartNode],
+        [parallelBodyId, parallelBodyNode],
+        [parallelEndId, parallelEndNode],
+        [afterParallelId, afterParallelNode],
+        [otherBranchId, otherBranchNode],
+      ])
+
+      const dag = createMockDAG(nodes)
+      const edgeManager = new EdgeManager(dag)
+
+      // Condition selects "else" branch
+      const readyNodes = edgeManager.processOutgoingEdges(conditionNode, { selectedOption: 'else' })
+
+      expect(readyNodes).toContain(otherBranchId)
+      expect(readyNodes).not.toContain(parallelStartId)
+      expect(readyNodes).not.toContain(afterParallelId)
+      // isNodeReady returns true when all edges are deactivated (no pending deps)
+      expect(edgeManager.isNodeReady(afterParallelNode)).toBe(true)
+    })
+
+    it('should still correctly handle normal loop exit (not deactivate when loop runs)', () => {
+      // When a loop actually executes and exits normally, after_loop should become ready
+      const sentinelStartId = 'sentinel-start'
+      const loopBodyId = 'loop-body'
+      const sentinelEndId = 'sentinel-end'
+      const afterLoopId = 'after-loop'
+
+      const sentinelStartNode = createMockNode(sentinelStartId, [{ target: loopBodyId }])
+
+      const loopBodyNode = createMockNode(
+        loopBodyId,
+        [{ target: sentinelEndId }],
+        [sentinelStartId]
+      )
+
+      const sentinelEndNode = createMockNode(
+        sentinelEndId,
+        [
+          { target: sentinelStartId, sourceHandle: 'loop_continue' },
+          { target: afterLoopId, sourceHandle: 'loop_exit' },
+        ],
+        [loopBodyId]
+      )
+
+      const afterLoopNode = createMockNode(afterLoopId, [], [sentinelEndId])
+
+      const nodes = new Map<string, DAGNode>([
+        [sentinelStartId, sentinelStartNode],
+        [loopBodyId, loopBodyNode],
+        [sentinelEndId, sentinelEndNode],
+        [afterLoopId, afterLoopNode],
+      ])
+
+      const dag = createMockDAG(nodes)
+      const edgeManager = new EdgeManager(dag)
+
+      // Simulate sentinel_end completing with loop_exit (loop is done)
+      const readyNodes = edgeManager.processOutgoingEdges(sentinelEndNode, {
+        selectedRoute: 'loop_exit',
+      })
+
+      // afterLoop should be ready
+      expect(readyNodes).toContain(afterLoopId)
+    })
+  })
 })

apps/sim/executor/execution/edge-manager.ts

@@ -243,7 +243,7 @@ export class EdgeManager {
     }
 
     for (const [, outgoingEdge] of targetNode.outgoingEdges) {
-      if (!this.isControlEdge(outgoingEdge.sourceHandle)) {
+      if (!this.isBackwardsEdge(outgoingEdge.sourceHandle)) {
         this.deactivateEdgeAndDescendants(
           targetId,
           outgoingEdge.target,

apps/sim/executor/handlers/agent/agent-handler.ts

@@ -25,6 +25,8 @@ import {
   validateModelProvider,
 } from '@/executor/utils/permission-check'
 import { executeProviderRequest } from '@/providers'
+import { transformAttachmentMessages } from '@/providers/attachment'
+import type { ProviderId } from '@/providers/types'
 import { getProviderFromModel, transformBlockTool } from '@/providers/utils'
 import type { SerializedBlock } from '@/serializer/types'
 import { executeTool } from '@/tools'
@@ -58,7 +60,15 @@ export class AgentBlockHandler implements BlockHandler {
     const providerId = getProviderFromModel(model)
     const formattedTools = await this.formatTools(ctx, filteredInputs.tools || [])
     const streamingConfig = this.getStreamingConfig(ctx, block)
-    const messages = await this.buildMessages(ctx, filteredInputs)
+    const rawMessages = await this.buildMessages(ctx, filteredInputs)
+
+    // Transform attachment messages to provider-specific format (async for file fetching)
+    const messages = rawMessages
+      ? await transformAttachmentMessages(rawMessages, {
+          providerId: providerId as ProviderId,
+          model,
+        })
+      : undefined
 
     const providerRequest = this.buildProviderRequest({
       ctx,
@@ -806,17 +816,44 @@ export class AgentBlockHandler implements BlockHandler {
     return messages.length > 0 ? messages : undefined
   }
 
-  private extractValidMessages(messages?: Message[]): Message[] {
-    if (!messages || !Array.isArray(messages)) return []
+  private extractValidMessages(messages?: Message[] | string): Message[] {
+    if (!messages) return []
 
-    return messages.filter(
-      (msg): msg is Message =>
-        msg &&
-        typeof msg === 'object' &&
-        'role' in msg &&
-        'content' in msg &&
-        ['system', 'user', 'assistant'].includes(msg.role)
-    )
+    // Handle raw JSON string input (from advanced mode)
+    let messageArray: unknown[]
+    if (typeof messages === 'string') {
+      const trimmed = messages.trim()
+      if (!trimmed) return []
+      try {
+        const parsed = JSON.parse(trimmed)
+        if (!Array.isArray(parsed)) {
+          logger.warn('Parsed messages JSON is not an array', { parsed })
+          return []
+        }
+        messageArray = parsed
+      } catch (error) {
+        logger.warn('Failed to parse messages JSON string', {
+          error,
+          messages: trimmed.substring(0, 100),
+        })
+        return []
+      }
+    } else if (Array.isArray(messages)) {
+      messageArray = messages
+    } else {
+      return []
+    }
+
+    return messageArray.filter((msg): msg is Message => {
+      if (!msg || typeof msg !== 'object') return false
+      const m = msg as Record<string, unknown>
+      return (
+        'role' in m &&
+        'content' in m &&
+        typeof m.role === 'string' &&
+        ['system', 'user', 'assistant', 'attachment'].includes(m.role)
+      )
+    })
   }
 
   private processMemories(memories: any): Message[] {

apps/sim/executor/handlers/agent/types.ts

@@ -6,8 +6,8 @@ export interface AgentInputs {
   systemPrompt?: string
   userPrompt?: string | object
   memories?: any // Legacy memory block output
-  // New message array input (from messages-input subblock)
-  messages?: Message[]
+  // New message array input (from messages-input subblock or raw JSON from advanced mode)
+  messages?: Message[] | string
   // Memory configuration
   memoryType?: 'none' | 'conversation' | 'sliding_window' | 'sliding_window_tokens'
   conversationId?: string // Required for all non-none memory types
@@ -42,9 +42,25 @@ export interface ToolInput {
   customToolId?: string
 }
 
+/**
+ * Attachment content (files, images, documents)
+ */
+export interface AttachmentContent {
+  /** Source type: how the data was provided */
+  sourceType: 'url' | 'base64' | 'file'
+  /** The URL or base64 data */
+  data: string
+  /** MIME type (e.g., 'image/png', 'application/pdf', 'audio/mp3') */
+  mimeType?: string
+  /** Optional filename for file uploads */
+  fileName?: string
+}
+
 export interface Message {
-  role: 'system' | 'user' | 'assistant'
+  role: 'system' | 'user' | 'assistant' | 'attachment'
   content: string
+  /** Attachment content for 'attachment' role messages */
+  attachment?: AttachmentContent
   executionId?: string
   function_call?: any
   tool_calls?: any[]

apps/sim/lib/copilot/orchestrator/config.ts

@@ -0,0 +1,42 @@
+/**
+ * Feature flag for server-side copilot orchestration.
+ */
+export const COPILOT_SERVER_ORCHESTRATED = true
+
+export const INTERRUPT_TOOL_NAMES = [
+  'set_global_workflow_variables',
+  'run_workflow',
+  'manage_mcp_tool',
+  'manage_custom_tool',
+  'deploy_mcp',
+  'deploy_chat',
+  'deploy_api',
+  'create_workspace_mcp_server',
+  'set_environment_variables',
+  'make_api_request',
+  'oauth_request_access',
+  'navigate_ui',
+  'knowledge_base',
+] as const
+
+export const INTERRUPT_TOOL_SET = new Set<string>(INTERRUPT_TOOL_NAMES)
+
+export const SUBAGENT_TOOL_NAMES = [
+  'debug',
+  'edit',
+  'plan',
+  'test',
+  'deploy',
+  'auth',
+  'research',
+  'knowledge',
+  'custom_tool',
+  'tour',
+  'info',
+  'workflow',
+  'evaluate',
+  'superagent',
+] as const
+
+export const SUBAGENT_TOOL_SET = new Set<string>(SUBAGENT_TOOL_NAMES)
+

apps/sim/lib/copilot/orchestrator/index.ts

@@ -0,0 +1,181 @@
+import { createLogger } from '@sim/logger'
+import { SIM_AGENT_API_URL_DEFAULT } from '@/lib/copilot/constants'
+import { env } from '@/lib/core/config/env'
+import { parseSSEStream } from '@/lib/copilot/orchestrator/sse-parser'
+import {
+  handleSubagentRouting,
+  sseHandlers,
+  subAgentHandlers,
+} from '@/lib/copilot/orchestrator/sse-handlers'
+import { prepareExecutionContext } from '@/lib/copilot/orchestrator/tool-executor'
+import type {
+  ExecutionContext,
+  OrchestratorOptions,
+  OrchestratorResult,
+  SSEEvent,
+  StreamingContext,
+  ToolCallSummary,
+} from '@/lib/copilot/orchestrator/types'
+
+const logger = createLogger('CopilotOrchestrator')
+const SIM_AGENT_API_URL = env.SIM_AGENT_API_URL || SIM_AGENT_API_URL_DEFAULT
+
+export interface OrchestrateStreamOptions extends OrchestratorOptions {
+  userId: string
+  workflowId: string
+  chatId?: string
+}
+
+/**
+ * Orchestrate a copilot SSE stream and execute tool calls server-side.
+ */
+export async function orchestrateCopilotStream(
+  requestPayload: Record<string, any>,
+  options: OrchestrateStreamOptions
+): Promise<OrchestratorResult> {
+  const { userId, workflowId, chatId, timeout = 300000, abortSignal } = options
+  const execContext = await prepareExecutionContext(userId, workflowId)
+
+  const context: StreamingContext = {
+    chatId,
+    conversationId: undefined,
+    messageId: requestPayload?.messageId || crypto.randomUUID(),
+    accumulatedContent: '',
+    contentBlocks: [],
+    toolCalls: new Map(),
+    currentThinkingBlock: null,
+    isInThinkingBlock: false,
+    subAgentParentToolCallId: undefined,
+    subAgentContent: {},
+    subAgentToolCalls: {},
+    pendingContent: '',
+    streamComplete: false,
+    wasAborted: false,
+    errors: [],
+  }
+
+  try {
+    const response = await fetch(`${SIM_AGENT_API_URL}/api/chat-completion-streaming`, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        ...(env.COPILOT_API_KEY ? { 'x-api-key': env.COPILOT_API_KEY } : {}),
+      },
+      body: JSON.stringify(requestPayload),
+      signal: abortSignal,
+    })
+
+    if (!response.ok) {
+      const errorText = await response.text().catch(() => '')
+      throw new Error(`Copilot backend error (${response.status}): ${errorText || response.statusText}`)
+    }
+
+    if (!response.body) {
+      throw new Error('Copilot backend response missing body')
+    }
+
+    const reader = response.body.getReader()
+    const decoder = new TextDecoder()
+
+    const timeoutId = setTimeout(() => {
+      context.errors.push('Request timed out')
+      context.streamComplete = true
+      reader.cancel().catch(() => {})
+    }, timeout)
+
+    try {
+      for await (const event of parseSSEStream(reader, decoder, abortSignal)) {
+        if (abortSignal?.aborted) {
+          context.wasAborted = true
+          break
+        }
+
+        await forwardEvent(event, options)
+
+        if (event.type === 'subagent_start') {
+          const toolCallId = event.data?.tool_call_id
+          if (toolCallId) {
+            context.subAgentParentToolCallId = toolCallId
+            context.subAgentContent[toolCallId] = ''
+            context.subAgentToolCalls[toolCallId] = []
+          }
+          continue
+        }
+
+        if (event.type === 'subagent_end') {
+          context.subAgentParentToolCallId = undefined
+          continue
+        }
+
+        if (handleSubagentRouting(event, context)) {
+          const handler = subAgentHandlers[event.type]
+          if (handler) {
+            await handler(event, context, execContext, options)
+          }
+          if (context.streamComplete) break
+          continue
+        }
+
+        const handler = sseHandlers[event.type]
+        if (handler) {
+          await handler(event, context, execContext, options)
+        }
+        if (context.streamComplete) break
+      }
+    } finally {
+      clearTimeout(timeoutId)
+    }
+
+    const result = buildResult(context)
+    await options.onComplete?.(result)
+    return result
+  } catch (error) {
+    const err = error instanceof Error ? error : new Error('Copilot orchestration failed')
+    logger.error('Copilot orchestration failed', { error: err.message })
+    await options.onError?.(err)
+    return {
+      success: false,
+      content: '',
+      contentBlocks: [],
+      toolCalls: [],
+      chatId: context.chatId,
+      conversationId: context.conversationId,
+      error: err.message,
+    }
+  }
+}
+
+async function forwardEvent(event: SSEEvent, options: OrchestratorOptions): Promise<void> {
+  try {
+    await options.onEvent?.(event)
+  } catch (error) {
+    logger.warn('Failed to forward SSE event', {
+      type: event.type,
+      error: error instanceof Error ? error.message : String(error),
+    })
+  }
+}
+
+function buildResult(context: StreamingContext): OrchestratorResult {
+  const toolCalls: ToolCallSummary[] = Array.from(context.toolCalls.values()).map((toolCall) => ({
+    id: toolCall.id,
+    name: toolCall.name,
+    status: toolCall.status,
+    params: toolCall.params,
+    result: toolCall.result?.output,
+    error: toolCall.error,
+    durationMs:
+      toolCall.endTime && toolCall.startTime ? toolCall.endTime - toolCall.startTime : undefined,
+  }))
+
+  return {
+    success: context.errors.length === 0,
+    content: context.accumulatedContent,
+    contentBlocks: context.contentBlocks,
+    toolCalls,
+    chatId: context.chatId,
+    conversationId: context.conversationId,
+    errors: context.errors.length ? context.errors : undefined,
+  }
+}
+

apps/sim/lib/copilot/orchestrator/persistence.ts

@@ -0,0 +1,138 @@
+import { db } from '@sim/db'
+import { copilotChats } from '@sim/db/schema'
+import { createLogger } from '@sim/logger'
+import { and, eq } from 'drizzle-orm'
+import { getRedisClient } from '@/lib/core/config/redis'
+
+const logger = createLogger('CopilotOrchestratorPersistence')
+
+/**
+ * Create a new copilot chat record.
+ */
+export async function createChat(params: {
+  userId: string
+  workflowId: string
+  model: string
+}): Promise<{ id: string }> {
+  const [chat] = await db
+    .insert(copilotChats)
+    .values({
+      userId: params.userId,
+      workflowId: params.workflowId,
+      model: params.model,
+      messages: [],
+    })
+    .returning({ id: copilotChats.id })
+
+  return { id: chat.id }
+}
+
+/**
+ * Load an existing chat for a user.
+ */
+export async function loadChat(chatId: string, userId: string) {
+  const [chat] = await db
+    .select()
+    .from(copilotChats)
+    .where(and(eq(copilotChats.id, chatId), eq(copilotChats.userId, userId)))
+    .limit(1)
+
+  return chat || null
+}
+
+/**
+ * Save chat messages and metadata.
+ */
+export async function saveMessages(
+  chatId: string,
+  messages: any[],
+  options?: {
+    title?: string
+    conversationId?: string
+    planArtifact?: string | null
+    config?: { mode?: string; model?: string }
+  }
+): Promise<void> {
+  await db
+    .update(copilotChats)
+    .set({
+      messages,
+      updatedAt: new Date(),
+      ...(options?.title ? { title: options.title } : {}),
+      ...(options?.conversationId ? { conversationId: options.conversationId } : {}),
+      ...(options?.planArtifact !== undefined ? { planArtifact: options.planArtifact } : {}),
+      ...(options?.config ? { config: options.config } : {}),
+    })
+    .where(eq(copilotChats.id, chatId))
+}
+
+/**
+ * Update the conversationId for a chat without overwriting messages.
+ */
+export async function updateChatConversationId(chatId: string, conversationId: string): Promise<void> {
+  await db
+    .update(copilotChats)
+    .set({
+      conversationId,
+      updatedAt: new Date(),
+    })
+    .where(eq(copilotChats.id, chatId))
+}
+
+/**
+ * Set a tool call confirmation status in Redis.
+ */
+export async function setToolConfirmation(
+  toolCallId: string,
+  status: 'accepted' | 'rejected' | 'background' | 'pending',
+  message?: string
+): Promise<boolean> {
+  const redis = getRedisClient()
+  if (!redis) {
+    logger.warn('Redis client not available for tool confirmation')
+    return false
+  }
+
+  const key = `tool_call:${toolCallId}`
+  const payload = {
+    status,
+    message: message || null,
+    timestamp: new Date().toISOString(),
+  }
+
+  try {
+    await redis.set(key, JSON.stringify(payload), 'EX', 86400)
+    return true
+  } catch (error) {
+    logger.error('Failed to set tool confirmation', {
+      toolCallId,
+      error: error instanceof Error ? error.message : String(error),
+    })
+    return false
+  }
+}
+
+/**
+ * Get a tool call confirmation status from Redis.
+ */
+export async function getToolConfirmation(toolCallId: string): Promise<{
+  status: string
+  message?: string
+  timestamp?: string
+} | null> {
+  const redis = getRedisClient()
+  if (!redis) return null
+
+  try {
+    const data = await redis.get(`tool_call:${toolCallId}`)
+    if (!data) return null
+    return JSON.parse(data) as { status: string; message?: string; timestamp?: string }
+  } catch (error) {
+    logger.error('Failed to read tool confirmation', {
+      toolCallId,
+      error: error instanceof Error ? error.message : String(error),
+    })
+    return null
+  }
+}
+

apps/sim/lib/copilot/orchestrator/sse-handlers.ts

@@ -0,0 +1,424 @@
+import { createLogger } from '@sim/logger'
+import type {
+  ContentBlock,
+  ExecutionContext,
+  OrchestratorOptions,
+  SSEEvent,
+  StreamingContext,
+  ToolCallState,
+} from '@/lib/copilot/orchestrator/types'
+import { executeToolServerSide, markToolComplete } from '@/lib/copilot/orchestrator/tool-executor'
+import { getToolConfirmation } from '@/lib/copilot/orchestrator/persistence'
+import { INTERRUPT_TOOL_SET, SUBAGENT_TOOL_SET } from '@/lib/copilot/orchestrator/config'
+
+const logger = createLogger('CopilotSseHandlers')
+
+/**
+ * Respond tools are internal to the copilot's subagent system.
+ * They're used by subagents to signal completion and should NOT be executed by the sim side.
+ * The copilot backend handles these internally.
+ */
+const RESPOND_TOOL_SET = new Set([
+  'plan_respond',
+  'edit_respond',
+  'debug_respond',
+  'info_respond',
+  'research_respond',
+  'deploy_respond',
+  'superagent_respond',
+])
+
+export type SSEHandler = (
+  event: SSEEvent,
+  context: StreamingContext,
+  execContext: ExecutionContext,
+  options: OrchestratorOptions
+) => void | Promise<void>
+
+function addContentBlock(
+  context: StreamingContext,
+  block: Omit<ContentBlock, 'timestamp'>
+): void {
+  context.contentBlocks.push({
+    ...block,
+    timestamp: Date.now(),
+  })
+}
+
+async function executeToolAndReport(
+  toolCallId: string,
+  context: StreamingContext,
+  execContext: ExecutionContext,
+  options?: OrchestratorOptions
+): Promise<void> {
+  const toolCall = context.toolCalls.get(toolCallId)
+  if (!toolCall) return
+
+  if (toolCall.status === 'executing') return
+
+  toolCall.status = 'executing'
+  try {
+    const result = await executeToolServerSide(toolCall, execContext)
+    toolCall.status = result.success ? 'success' : 'error'
+    toolCall.result = result
+    toolCall.error = result.error
+    toolCall.endTime = Date.now()
+
+    await markToolComplete(
+      toolCall.id,
+      toolCall.name,
+      result.success ? 200 : 500,
+      result.error || (result.success ? 'Tool completed' : 'Tool failed'),
+      result.output
+    )
+
+    await options?.onEvent?.({
+      type: 'tool_result',
+      toolCallId: toolCall.id,
+      data: {
+        id: toolCall.id,
+        name: toolCall.name,
+        success: result.success,
+        result: result.output,
+      },
+    })
+  } catch (error) {
+    toolCall.status = 'error'
+    toolCall.error = error instanceof Error ? error.message : String(error)
+    toolCall.endTime = Date.now()
+
+    await markToolComplete(toolCall.id, toolCall.name, 500, toolCall.error)
+
+    await options?.onEvent?.({
+      type: 'tool_error',
+      toolCallId: toolCall.id,
+      data: {
+        id: toolCall.id,
+        name: toolCall.name,
+        error: toolCall.error,
+      },
+    })
+  }
+}
+
+async function waitForToolDecision(
+  toolCallId: string,
+  timeoutMs: number
+): Promise<{ status: string; message?: string } | null> {
+  const start = Date.now()
+  while (Date.now() - start < timeoutMs) {
+    const decision = await getToolConfirmation(toolCallId)
+    if (decision?.status) {
+      return decision
+    }
+    await new Promise((resolve) => setTimeout(resolve, 100))
+  }
+  return null
+}
+
+export const sseHandlers: Record<string, SSEHandler> = {
+  chat_id: (event, context) => {
+    context.chatId = event.data?.chatId
+  },
+  title_updated: () => {},
+  tool_result: (event, context) => {
+    const toolCallId = event.toolCallId || event.data?.id
+    if (!toolCallId) return
+    const current = context.toolCalls.get(toolCallId)
+    if (!current) return
+
+    // Determine success: explicit success field, or if there's result data without explicit failure
+    const hasExplicitSuccess = event.data?.success !== undefined || event.data?.result?.success !== undefined
+    const explicitSuccess = event.data?.success ?? event.data?.result?.success
+    const hasResultData = event.data?.result !== undefined || event.data?.data !== undefined
+    const hasError = !!event.data?.error || !!event.data?.result?.error
+
+    // If explicitly set, use that; otherwise infer from data presence
+    const success = hasExplicitSuccess ? !!explicitSuccess : (hasResultData && !hasError)
+
+    current.status = success ? 'success' : 'error'
+    current.endTime = Date.now()
+    if (hasResultData) {
+      current.result = {
+        success,
+        output: event.data?.result || event.data?.data,
+      }
+    }
+    if (hasError) {
+      current.error = event.data?.error || event.data?.result?.error
+    }
+  },
+  tool_error: (event, context) => {
+    const toolCallId = event.toolCallId || event.data?.id
+    if (!toolCallId) return
+    const current = context.toolCalls.get(toolCallId)
+    if (!current) return
+    current.status = 'error'
+    current.error = event.data?.error || 'Tool execution failed'
+    current.endTime = Date.now()
+  },
+  tool_generating: (event, context) => {
+    const toolCallId = event.toolCallId || event.data?.toolCallId || event.data?.id
+    const toolName = event.toolName || event.data?.toolName || event.data?.name
+    if (!toolCallId || !toolName) return
+    if (!context.toolCalls.has(toolCallId)) {
+      context.toolCalls.set(toolCallId, {
+        id: toolCallId,
+        name: toolName,
+        status: 'pending',
+        startTime: Date.now(),
+      })
+    }
+  },
+  tool_call: async (event, context, execContext, options) => {
+    const toolData = event.data || {}
+    const toolCallId = toolData.id || event.toolCallId
+    const toolName = toolData.name || event.toolName
+    if (!toolCallId || !toolName) return
+
+    const args = toolData.arguments || toolData.input || event.data?.input
+    const isPartial = toolData.partial === true
+    const existing = context.toolCalls.get(toolCallId)
+    const toolCall: ToolCallState = existing
+      ? { ...existing, status: 'pending', params: args || existing.params }
+      : {
+          id: toolCallId,
+          name: toolName,
+          status: 'pending',
+          params: args,
+          startTime: Date.now(),
+        }
+
+    context.toolCalls.set(toolCallId, toolCall)
+    addContentBlock(context, { type: 'tool_call', toolCall })
+
+    if (isPartial) return
+
+    // Subagent tools are executed by the copilot backend, not sim side
+    if (SUBAGENT_TOOL_SET.has(toolName)) {
+      return
+    }
+
+    // Respond tools are internal to copilot's subagent system - skip execution
+    // The copilot backend handles these internally to signal subagent completion
+    if (RESPOND_TOOL_SET.has(toolName)) {
+      toolCall.status = 'success'
+      toolCall.endTime = Date.now()
+      toolCall.result = { success: true, output: 'Internal respond tool - handled by copilot backend' }
+      return
+    }
+
+    const isInterruptTool = INTERRUPT_TOOL_SET.has(toolName)
+    const isInteractive = options.interactive === true
+
+    if (isInterruptTool && isInteractive) {
+      const decision = await waitForToolDecision(toolCallId, options.timeout || 600000)
+      if (decision?.status === 'accepted' || decision?.status === 'success') {
+        await executeToolAndReport(toolCallId, context, execContext, options)
+        return
+      }
+
+      if (decision?.status === 'rejected' || decision?.status === 'error') {
+        toolCall.status = 'rejected'
+        toolCall.endTime = Date.now()
+        await markToolComplete(
+          toolCall.id,
+          toolCall.name,
+          400,
+          decision.message || 'Tool execution rejected',
+          { skipped: true, reason: 'user_rejected' }
+        )
+        await options.onEvent?.({
+          type: 'tool_result',
+          toolCallId: toolCall.id,
+          data: {
+            id: toolCall.id,
+            name: toolCall.name,
+            success: false,
+            result: { skipped: true, reason: 'user_rejected' },
+          },
+        })
+        return
+      }
+
+      if (decision?.status === 'background') {
+        toolCall.status = 'skipped'
+        toolCall.endTime = Date.now()
+        await markToolComplete(
+          toolCall.id,
+          toolCall.name,
+          202,
+          decision.message || 'Tool execution moved to background',
+          { background: true }
+        )
+        await options.onEvent?.({
+          type: 'tool_result',
+          toolCallId: toolCall.id,
+          data: {
+            id: toolCall.id,
+            name: toolCall.name,
+            success: true,
+            result: { background: true },
+          },
+        })
+        return
+      }
+    }
+
+    if (options.autoExecuteTools !== false) {
+      await executeToolAndReport(toolCallId, context, execContext, options)
+    }
+  },
+  reasoning: (event, context) => {
+    const phase = event.data?.phase || event.data?.data?.phase
+    if (phase === 'start') {
+      context.isInThinkingBlock = true
+      context.currentThinkingBlock = {
+        type: 'thinking',
+        content: '',
+        timestamp: Date.now(),
+      }
+      return
+    }
+    if (phase === 'end') {
+      if (context.currentThinkingBlock) {
+        context.contentBlocks.push(context.currentThinkingBlock)
+      }
+      context.isInThinkingBlock = false
+      context.currentThinkingBlock = null
+      return
+    }
+    const chunk = typeof event.data === 'string' ? event.data : event.data?.data || event.data?.content
+    if (!chunk || !context.currentThinkingBlock) return
+    context.currentThinkingBlock.content = `${context.currentThinkingBlock.content || ''}${chunk}`
+  },
+  content: (event, context) => {
+    const chunk = typeof event.data === 'string' ? event.data : event.data?.content || event.data?.data
+    if (!chunk) return
+    context.accumulatedContent += chunk
+    addContentBlock(context, { type: 'text', content: chunk })
+  },
+  done: (event, context) => {
+    if (event.data?.responseId) {
+      context.conversationId = event.data.responseId
+    }
+    context.streamComplete = true
+  },
+  start: (event, context) => {
+    if (event.data?.responseId) {
+      context.conversationId = event.data.responseId
+    }
+  },
+  error: (event, context) => {
+    const message =
+      event.data?.message || event.data?.error || (typeof event.data === 'string' ? event.data : null)
+    if (message) {
+      context.errors.push(message)
+    }
+    context.streamComplete = true
+  },
+}
+
+export const subAgentHandlers: Record<string, SSEHandler> = {
+  content: (event, context) => {
+    const parentToolCallId = context.subAgentParentToolCallId
+    if (!parentToolCallId || !event.data) return
+    const chunk = typeof event.data === 'string' ? event.data : event.data?.content || ''
+    if (!chunk) return
+    context.subAgentContent[parentToolCallId] = (context.subAgentContent[parentToolCallId] || '') + chunk
+    addContentBlock(context, { type: 'subagent_text', content: chunk })
+  },
+  tool_call: async (event, context, execContext, options) => {
+    const parentToolCallId = context.subAgentParentToolCallId
+    if (!parentToolCallId) return
+    const toolData = event.data || {}
+    const toolCallId = toolData.id || event.toolCallId
+    const toolName = toolData.name || event.toolName
+    if (!toolCallId || !toolName) return
+    const isPartial = toolData.partial === true
+    const args = toolData.arguments || toolData.input || event.data?.input
+
+    const toolCall: ToolCallState = {
+      id: toolCallId,
+      name: toolName,
+      status: 'pending',
+      params: args,
+      startTime: Date.now(),
+    }
+
+    // Store in both places - subAgentToolCalls for tracking and toolCalls for executeToolAndReport
+    if (!context.subAgentToolCalls[parentToolCallId]) {
+      context.subAgentToolCalls[parentToolCallId] = []
+    }
+    context.subAgentToolCalls[parentToolCallId].push(toolCall)
+    context.toolCalls.set(toolCallId, toolCall)
+
+    if (isPartial) return
+
+    // Respond tools are internal to copilot's subagent system - skip execution
+    if (RESPOND_TOOL_SET.has(toolName)) {
+      toolCall.status = 'success'
+      toolCall.endTime = Date.now()
+      toolCall.result = { success: true, output: 'Internal respond tool - handled by copilot backend' }
+      return
+    }
+
+    if (options.autoExecuteTools !== false) {
+      await executeToolAndReport(toolCallId, context, execContext, options)
+    }
+  },
+  tool_result: (event, context) => {
+    const parentToolCallId = context.subAgentParentToolCallId
+    if (!parentToolCallId) return
+    const toolCallId = event.toolCallId || event.data?.id
+    if (!toolCallId) return
+
+    // Update in subAgentToolCalls
+    const toolCalls = context.subAgentToolCalls[parentToolCallId] || []
+    const subAgentToolCall = toolCalls.find((tc) => tc.id === toolCallId)
+
+    // Also update in main toolCalls (where we added it for execution)
+    const mainToolCall = context.toolCalls.get(toolCallId)
+
+    // Use same success inference logic as main handler
+    const hasExplicitSuccess =
+      event.data?.success !== undefined || event.data?.result?.success !== undefined
+    const explicitSuccess = event.data?.success ?? event.data?.result?.success
+    const hasResultData = event.data?.result !== undefined || event.data?.data !== undefined
+    const hasError = !!event.data?.error || !!event.data?.result?.error
+    const success = hasExplicitSuccess ? !!explicitSuccess : hasResultData && !hasError
+
+    const status = success ? 'success' : 'error'
+    const endTime = Date.now()
+    const result = hasResultData
+      ? { success, output: event.data?.result || event.data?.data }
+      : undefined
+
+    if (subAgentToolCall) {
+      subAgentToolCall.status = status
+      subAgentToolCall.endTime = endTime
+      if (result) subAgentToolCall.result = result
+      if (hasError) subAgentToolCall.error = event.data?.error || event.data?.result?.error
+    }
+
+    if (mainToolCall) {
+      mainToolCall.status = status
+      mainToolCall.endTime = endTime
+      if (result) mainToolCall.result = result
+      if (hasError) mainToolCall.error = event.data?.error || event.data?.result?.error
+    }
+  },
+}
+
+export function handleSubagentRouting(event: SSEEvent, context: StreamingContext): boolean {
+  if (!event.subagent) return false
+  if (!context.subAgentParentToolCallId) {
+    logger.warn('Subagent event missing parent tool call', {
+      type: event.type,
+      subagent: event.subagent,
+    })
+    return false
+  }
+  return true
+}
+

apps/sim/lib/copilot/orchestrator/sse-parser.ts

@@ -0,0 +1,72 @@
+import { createLogger } from '@sim/logger'
+import type { SSEEvent } from '@/lib/copilot/orchestrator/types'
+
+const logger = createLogger('CopilotSseParser')
+
+/**
+ * Parses SSE streams from the copilot backend into typed events.
+ */
+export async function* parseSSEStream(
+  reader: ReadableStreamDefaultReader<Uint8Array>,
+  decoder: TextDecoder,
+  abortSignal?: AbortSignal
+): AsyncGenerator<SSEEvent> {
+  let buffer = ''
+
+  try {
+    while (true) {
+      if (abortSignal?.aborted) {
+        logger.info('SSE stream aborted by signal')
+        break
+      }
+
+      const { done, value } = await reader.read()
+      if (done) break
+
+      buffer += decoder.decode(value, { stream: true })
+      const lines = buffer.split('\n')
+      buffer = lines.pop() || ''
+
+      for (const line of lines) {
+        if (!line.trim()) continue
+        if (!line.startsWith('data: ')) continue
+
+        const jsonStr = line.slice(6)
+        if (jsonStr === '[DONE]') continue
+
+        try {
+          const event = JSON.parse(jsonStr) as SSEEvent
+          if (event?.type) {
+            yield event
+          }
+        } catch (error) {
+          logger.warn('Failed to parse SSE event', {
+            preview: jsonStr.slice(0, 200),
+            error: error instanceof Error ? error.message : String(error),
+          })
+        }
+      }
+    }
+
+    if (buffer.trim() && buffer.startsWith('data: ')) {
+      try {
+        const event = JSON.parse(buffer.slice(6)) as SSEEvent
+        if (event?.type) {
+          yield event
+        }
+      } catch (error) {
+        logger.warn('Failed to parse final SSE buffer', {
+          preview: buffer.slice(0, 200),
+          error: error instanceof Error ? error.message : String(error),
+        })
+      }
+    }
+  } finally {
+    try {
+      reader.releaseLock()
+    } catch {
+      logger.warn('Failed to release SSE reader lock')
+    }
+  }
+}
+

apps/sim/lib/copilot/orchestrator/types.ts

@@ -0,0 +1,127 @@
+import type { CopilotProviderConfig } from '@/lib/copilot/types'
+
+export type SSEEventType =
+  | 'chat_id'
+  | 'title_updated'
+  | 'content'
+  | 'reasoning'
+  | 'tool_call'
+  | 'tool_generating'
+  | 'tool_result'
+  | 'tool_error'
+  | 'subagent_start'
+  | 'subagent_end'
+  | 'done'
+  | 'error'
+  | 'start'
+
+export interface SSEEvent {
+  type: SSEEventType
+  data?: any
+  subagent?: string
+  toolCallId?: string
+  toolName?: string
+}
+
+export type ToolCallStatus = 'pending' | 'executing' | 'success' | 'error' | 'skipped' | 'rejected'
+
+export interface ToolCallState {
+  id: string
+  name: string
+  status: ToolCallStatus
+  params?: Record<string, any>
+  result?: ToolCallResult
+  error?: string
+  startTime?: number
+  endTime?: number
+}
+
+export interface ToolCallResult {
+  success: boolean
+  output?: any
+  error?: string
+}
+
+export type ContentBlockType = 'text' | 'thinking' | 'tool_call' | 'subagent_text'
+
+export interface ContentBlock {
+  type: ContentBlockType
+  content?: string
+  toolCall?: ToolCallState
+  timestamp: number
+}
+
+export interface StreamingContext {
+  chatId?: string
+  conversationId?: string
+  messageId: string
+  accumulatedContent: string
+  contentBlocks: ContentBlock[]
+  toolCalls: Map<string, ToolCallState>
+  currentThinkingBlock: ContentBlock | null
+  isInThinkingBlock: boolean
+  subAgentParentToolCallId?: string
+  subAgentContent: Record<string, string>
+  subAgentToolCalls: Record<string, ToolCallState[]>
+  pendingContent: string
+  streamComplete: boolean
+  wasAborted: boolean
+  errors: string[]
+}
+
+export interface OrchestratorRequest {
+  message: string
+  workflowId: string
+  userId: string
+  chatId?: string
+  mode?: 'agent' | 'ask' | 'plan'
+  model?: string
+  conversationId?: string
+  contexts?: Array<{ type: string; content: string }>
+  fileAttachments?: any[]
+  commands?: string[]
+  provider?: CopilotProviderConfig
+  streamToolCalls?: boolean
+  version?: string
+  prefetch?: boolean
+  userName?: string
+}
+
+export interface OrchestratorOptions {
+  autoExecuteTools?: boolean
+  timeout?: number
+  onEvent?: (event: SSEEvent) => void | Promise<void>
+  onComplete?: (result: OrchestratorResult) => void | Promise<void>
+  onError?: (error: Error) => void | Promise<void>
+  abortSignal?: AbortSignal
+  interactive?: boolean
+}
+
+export interface OrchestratorResult {
+  success: boolean
+  content: string
+  contentBlocks: ContentBlock[]
+  toolCalls: ToolCallSummary[]
+  chatId?: string
+  conversationId?: string
+  error?: string
+  errors?: string[]
+}
+
+export interface ToolCallSummary {
+  id: string
+  name: string
+  status: ToolCallStatus
+  params?: Record<string, any>
+  result?: any
+  error?: string
+  durationMs?: number
+}
+
+export interface ExecutionContext {
+  userId: string
+  workflowId: string
+  workspaceId?: string
+  decryptedEnvVars?: Record<string, string>
+}
+

apps/sim/lib/copilot/tools/client/knowledge/knowledge-base.ts

@@ -10,6 +10,7 @@ import {
   type KnowledgeBaseArgs,
 } from '@/lib/copilot/tools/shared/schemas'
 import { useCopilotStore } from '@/stores/panel/copilot/store'
+import { useWorkflowRegistry } from '@/stores/workflows/registry/store'
 
 /**
  * Client tool for knowledge base operations
@@ -102,7 +103,19 @@ export class KnowledgeBaseClientTool extends BaseClientTool {
     const logger = createLogger('KnowledgeBaseClientTool')
     try {
       this.setState(ClientToolCallState.executing)
-      const payload: KnowledgeBaseArgs = { ...(args || { operation: 'list' }) }
+
+      // Get the workspace ID from the workflow registry hydration state
+      const { hydration } = useWorkflowRegistry.getState()
+      const workspaceId = hydration.workspaceId
+
+      // Build payload with workspace ID included in args
+      const payload: KnowledgeBaseArgs = {
+        ...(args || { operation: 'list' }),
+        args: {
+          ...(args?.args || {}),
+          workspaceId: workspaceId || undefined,
+        },
+      }
 
       const res = await fetch('/api/copilot/execute-copilot-server-tool', {
         method: 'POST',

apps/sim/lib/copilot/tools/server/workflow/edit-workflow.ts

@@ -8,7 +8,11 @@ import { validateSelectorIds } from '@/lib/copilot/validation/selector-validator
 import type { PermissionGroupConfig } from '@/lib/permission-groups/types'
 import { getBlockOutputs } from '@/lib/workflows/blocks/block-outputs'
 import { extractAndPersistCustomTools } from '@/lib/workflows/persistence/custom-tools-persistence'
-import { loadWorkflowFromNormalizedTables } from '@/lib/workflows/persistence/utils'
+import { applyAutoLayout } from '@/lib/workflows/autolayout'
+import {
+  loadWorkflowFromNormalizedTables,
+  saveWorkflowToNormalizedTables,
+} from '@/lib/workflows/persistence/utils'
 import { isValidKey } from '@/lib/workflows/sanitization/key-validation'
 import { validateWorkflowState } from '@/lib/workflows/sanitization/validation'
 import { buildCanonicalIndex, isCanonicalPair } from '@/lib/workflows/subblocks/visibility'
@@ -2508,6 +2512,10 @@ async function validateWorkflowSelectorIds(
     for (const subBlockConfig of blockConfig.subBlocks) {
       if (!SELECTOR_TYPES.has(subBlockConfig.type)) continue
 
+      // Skip oauth-input - credentials are pre-validated before edit application
+      // This allows existing collaborator credentials to remain untouched
+      if (subBlockConfig.type === 'oauth-input') continue
+
       const subBlockValue = blockData.subBlocks?.[subBlockConfig.id]?.value
       if (!subBlockValue) continue
 
@@ -2573,6 +2581,295 @@ async function validateWorkflowSelectorIds(
   return errors
 }
 
+/**
+ * Pre-validates credential and apiKey inputs in operations before they are applied.
+ * - Validates oauth-input (credential) IDs belong to the user
+ * - Filters out apiKey inputs for hosted models when isHosted is true
+ * - Also validates credentials and apiKeys in nestedNodes (blocks inside loop/parallel)
+ * Returns validation errors for any removed inputs.
+ */
+async function preValidateCredentialInputs(
+  operations: EditWorkflowOperation[],
+  context: { userId: string },
+  workflowState?: Record<string, unknown>
+): Promise<{ filteredOperations: EditWorkflowOperation[]; errors: ValidationError[] }> {
+  const { isHosted } = await import('@/lib/core/config/feature-flags')
+  const { getHostedModels } = await import('@/providers/utils')
+
+  const logger = createLogger('PreValidateCredentials')
+  const errors: ValidationError[] = []
+
+  // Collect credential and apiKey inputs that need validation/filtering
+  const credentialInputs: Array<{
+    operationIndex: number
+    blockId: string
+    blockType: string
+    fieldName: string
+    value: string
+    nestedBlockId?: string
+  }> = []
+
+  const hostedApiKeyInputs: Array<{
+    operationIndex: number
+    blockId: string
+    blockType: string
+    model: string
+    nestedBlockId?: string
+  }> = []
+
+  const hostedModelsLower = isHosted ? new Set(getHostedModels().map((m) => m.toLowerCase())) : null
+
+  /**
+   * Collect credential inputs from a block's inputs based on its block config
+   */
+  function collectCredentialInputs(
+    blockConfig: ReturnType<typeof getBlock>,
+    inputs: Record<string, unknown>,
+    opIndex: number,
+    blockId: string,
+    blockType: string,
+    nestedBlockId?: string
+  ) {
+    if (!blockConfig) return
+
+    for (const subBlockConfig of blockConfig.subBlocks) {
+      if (subBlockConfig.type !== 'oauth-input') continue
+
+      const inputValue = inputs[subBlockConfig.id]
+      if (!inputValue || typeof inputValue !== 'string' || inputValue.trim() === '') continue
+
+      credentialInputs.push({
+        operationIndex: opIndex,
+        blockId,
+        blockType,
+        fieldName: subBlockConfig.id,
+        value: inputValue,
+        nestedBlockId,
+      })
+    }
+  }
+
+  /**
+   * Check if apiKey should be filtered for a block with the given model
+   */
+  function collectHostedApiKeyInput(
+    inputs: Record<string, unknown>,
+    modelValue: string | undefined,
+    opIndex: number,
+    blockId: string,
+    blockType: string,
+    nestedBlockId?: string
+  ) {
+    if (!hostedModelsLower || !inputs.apiKey) return
+    if (!modelValue || typeof modelValue !== 'string') return
+
+    if (hostedModelsLower.has(modelValue.toLowerCase())) {
+      hostedApiKeyInputs.push({
+        operationIndex: opIndex,
+        blockId,
+        blockType,
+        model: modelValue,
+        nestedBlockId,
+      })
+    }
+  }
+
+  operations.forEach((op, opIndex) => {
+    // Process main block inputs
+    if (op.params?.inputs && op.params?.type) {
+      const blockConfig = getBlock(op.params.type)
+      if (blockConfig) {
+        // Collect credentials from main block
+        collectCredentialInputs(
+          blockConfig,
+          op.params.inputs as Record<string, unknown>,
+          opIndex,
+          op.block_id,
+          op.params.type
+        )
+
+        // Check for apiKey inputs on hosted models
+        let modelValue = (op.params.inputs as Record<string, unknown>).model as string | undefined
+
+        // For edit operations, if model is not being changed, check existing block's model
+        if (
+          !modelValue &&
+          op.operation_type === 'edit' &&
+          (op.params.inputs as Record<string, unknown>).apiKey &&
+          workflowState
+        ) {
+          const existingBlock = (workflowState.blocks as Record<string, unknown>)?.[op.block_id] as
+            | Record<string, unknown>
+            | undefined
+          const existingSubBlocks = existingBlock?.subBlocks as Record<string, unknown> | undefined
+          const existingModelSubBlock = existingSubBlocks?.model as
+            | Record<string, unknown>
+            | undefined
+          modelValue = existingModelSubBlock?.value as string | undefined
+        }
+
+        collectHostedApiKeyInput(
+          op.params.inputs as Record<string, unknown>,
+          modelValue,
+          opIndex,
+          op.block_id,
+          op.params.type
+        )
+      }
+    }
+
+    // Process nested nodes (blocks inside loop/parallel containers)
+    const nestedNodes = op.params?.nestedNodes as
+      | Record<string, Record<string, unknown>>
+      | undefined
+    if (nestedNodes) {
+      Object.entries(nestedNodes).forEach(([childId, childBlock]) => {
+        const childType = childBlock.type as string | undefined
+        const childInputs = childBlock.inputs as Record<string, unknown> | undefined
+        if (!childType || !childInputs) return
+
+        const childBlockConfig = getBlock(childType)
+        if (!childBlockConfig) return
+
+        // Collect credentials from nested block
+        collectCredentialInputs(
+          childBlockConfig,
+          childInputs,
+          opIndex,
+          op.block_id,
+          childType,
+          childId
+        )
+
+        // Check for apiKey inputs on hosted models in nested block
+        const modelValue = childInputs.model as string | undefined
+        collectHostedApiKeyInput(childInputs, modelValue, opIndex, op.block_id, childType, childId)
+      })
+    }
+  })
+
+  const hasCredentialsToValidate = credentialInputs.length > 0
+  const hasHostedApiKeysToFilter = hostedApiKeyInputs.length > 0
+
+  if (!hasCredentialsToValidate && !hasHostedApiKeysToFilter) {
+    return { filteredOperations: operations, errors }
+  }
+
+  // Deep clone operations so we can modify them
+  const filteredOperations = structuredClone(operations)
+
+  // Filter out apiKey inputs for hosted models and add validation errors
+  if (hasHostedApiKeysToFilter) {
+    logger.info('Filtering apiKey inputs for hosted models', { count: hostedApiKeyInputs.length })
+
+    for (const apiKeyInput of hostedApiKeyInputs) {
+      const op = filteredOperations[apiKeyInput.operationIndex]
+
+      // Handle nested block apiKey filtering
+      if (apiKeyInput.nestedBlockId) {
+        const nestedNodes = op.params?.nestedNodes as
+          | Record<string, Record<string, unknown>>
+          | undefined
+        const nestedBlock = nestedNodes?.[apiKeyInput.nestedBlockId]
+        const nestedInputs = nestedBlock?.inputs as Record<string, unknown> | undefined
+        if (nestedInputs?.apiKey) {
+          nestedInputs.apiKey = undefined
+          logger.debug('Filtered apiKey for hosted model in nested block', {
+            parentBlockId: apiKeyInput.blockId,
+            nestedBlockId: apiKeyInput.nestedBlockId,
+            model: apiKeyInput.model,
+          })
+
+          errors.push({
+            blockId: apiKeyInput.nestedBlockId,
+            blockType: apiKeyInput.blockType,
+            field: 'apiKey',
+            value: '[redacted]',
+            error: `Cannot set API key for hosted model "${apiKeyInput.model}" - API keys are managed by the platform when using hosted models`,
+          })
+        }
+      } else if (op.params?.inputs?.apiKey) {
+        // Handle main block apiKey filtering
+        op.params.inputs.apiKey = undefined
+        logger.debug('Filtered apiKey for hosted model', {
+          blockId: apiKeyInput.blockId,
+          model: apiKeyInput.model,
+        })
+
+        errors.push({
+          blockId: apiKeyInput.blockId,
+          blockType: apiKeyInput.blockType,
+          field: 'apiKey',
+          value: '[redacted]',
+          error: `Cannot set API key for hosted model "${apiKeyInput.model}" - API keys are managed by the platform when using hosted models`,
+        })
+      }
+    }
+  }
+
+  // Validate credential inputs
+  if (hasCredentialsToValidate) {
+    logger.info('Pre-validating credential inputs', {
+      credentialCount: credentialInputs.length,
+      userId: context.userId,
+    })
+
+    const allCredentialIds = credentialInputs.map((c) => c.value)
+    const validationResult = await validateSelectorIds('oauth-input', allCredentialIds, context)
+    const invalidSet = new Set(validationResult.invalid)
+
+    if (invalidSet.size > 0) {
+      for (const credInput of credentialInputs) {
+        if (!invalidSet.has(credInput.value)) continue
+
+        const op = filteredOperations[credInput.operationIndex]
+
+        // Handle nested block credential removal
+        if (credInput.nestedBlockId) {
+          const nestedNodes = op.params?.nestedNodes as
+            | Record<string, Record<string, unknown>>
+            | undefined
+          const nestedBlock = nestedNodes?.[credInput.nestedBlockId]
+          const nestedInputs = nestedBlock?.inputs as Record<string, unknown> | undefined
+          if (nestedInputs?.[credInput.fieldName]) {
+            delete nestedInputs[credInput.fieldName]
+            logger.info('Removed invalid credential from nested block', {
+              parentBlockId: credInput.blockId,
+              nestedBlockId: credInput.nestedBlockId,
+              field: credInput.fieldName,
+              invalidValue: credInput.value,
+            })
+          }
+        } else if (op.params?.inputs?.[credInput.fieldName]) {
+          // Handle main block credential removal
+          delete op.params.inputs[credInput.fieldName]
+          logger.info('Removed invalid credential from operation', {
+            blockId: credInput.blockId,
+            field: credInput.fieldName,
+            invalidValue: credInput.value,
+          })
+        }
+
+        const warningInfo = validationResult.warning ? `. ${validationResult.warning}` : ''
+        const errorBlockId = credInput.nestedBlockId ?? credInput.blockId
+        errors.push({
+          blockId: errorBlockId,
+          blockType: credInput.blockType,
+          field: credInput.fieldName,
+          value: credInput.value,
+          error: `Invalid credential ID "${credInput.value}" - credential does not exist or user doesn't have access${warningInfo}`,
+        })
+      }
+
+      logger.warn('Filtered out invalid credentials', {
+        invalidCount: invalidSet.size,
+      })
+    }
+  }
+
+  return { filteredOperations, errors }
+}
+
 async function getCurrentWorkflowStateFromDb(
   workflowId: string
 ): Promise<{ workflowState: any; subBlockValues: Record<string, Record<string, any>> }> {
@@ -2657,12 +2954,29 @@ export const editWorkflowServerTool: BaseServerTool<EditWorkflowParams, any> = {
     // Get permission config for the user
     const permissionConfig = context?.userId ? await getUserPermissionConfig(context.userId) : null
 
+    // Pre-validate credential and apiKey inputs before applying operations
+    // This filters out invalid credentials and apiKeys for hosted models
+    let operationsToApply = operations
+    const credentialErrors: ValidationError[] = []
+    if (context?.userId) {
+      const { filteredOperations, errors: credErrors } = await preValidateCredentialInputs(
+        operations,
+        { userId: context.userId },
+        workflowState
+      )
+      operationsToApply = filteredOperations
+      credentialErrors.push(...credErrors)
+    }
+
     // Apply operations directly to the workflow state
     const {
       state: modifiedWorkflowState,
       validationErrors,
       skippedItems,
-    } = applyOperationsToWorkflowState(workflowState, operations, permissionConfig)
+    } = applyOperationsToWorkflowState(workflowState, operationsToApply, permissionConfig)
+
+    // Add credential validation errors
+    validationErrors.push(...credentialErrors)
 
     // Get workspaceId for selector validation
     let workspaceId: string | undefined
@@ -2757,10 +3071,60 @@ export const editWorkflowServerTool: BaseServerTool<EditWorkflowParams, any> = {
     const skippedMessages =
       skippedItems.length > 0 ? skippedItems.map((item) => item.reason) : undefined
 
-    // Return the modified workflow state for the client to convert to YAML if needed
+    // Persist the workflow state to the database
+    const finalWorkflowState = validation.sanitizedState || modifiedWorkflowState
+
+    // Apply autolayout to position blocks properly
+    const layoutResult = applyAutoLayout(finalWorkflowState.blocks, finalWorkflowState.edges, {
+      horizontalSpacing: 250,
+      verticalSpacing: 100,
+      padding: { x: 100, y: 100 },
+    })
+
+    const layoutedBlocks = layoutResult.success && layoutResult.blocks
+      ? layoutResult.blocks
+      : finalWorkflowState.blocks
+
+    if (!layoutResult.success) {
+      logger.warn('Autolayout failed, using default positions', {
+        workflowId,
+        error: layoutResult.error,
+      })
+    }
+
+    const workflowStateForDb = {
+      blocks: layoutedBlocks,
+      edges: finalWorkflowState.edges,
+      loops: generateLoopBlocks(layoutedBlocks as any),
+      parallels: generateParallelBlocks(layoutedBlocks as any),
+      lastSaved: Date.now(),
+      isDeployed: false,
+    }
+
+    const saveResult = await saveWorkflowToNormalizedTables(workflowId, workflowStateForDb as any)
+    if (!saveResult.success) {
+      logger.error('Failed to persist workflow state to database', {
+        workflowId,
+        error: saveResult.error,
+      })
+      throw new Error(`Failed to save workflow: ${saveResult.error}`)
+    }
+
+    // Update workflow's lastSynced timestamp
+    await db
+      .update(workflowTable)
+      .set({
+        lastSynced: new Date(),
+        updatedAt: new Date(),
+      })
+      .where(eq(workflowTable.id, workflowId))
+
+    logger.info('Workflow state persisted to database', { workflowId })
+
+    // Return the modified workflow state with autolayout applied
     return {
       success: true,
-      workflowState: validation.sanitizedState || modifiedWorkflowState,
+      workflowState: { ...finalWorkflowState, blocks: layoutedBlocks },
       // Include input validation errors so the LLM can see what was rejected
       ...(inputErrors && {
         inputValidationErrors: inputErrors,

apps/sim/next.config.ts

@@ -325,18 +325,6 @@ const nextConfig: NextConfig = {
 
     return redirects
   },
-  async rewrites() {
-    return [
-      {
-        source: '/ingest/static/:path*',
-        destination: 'https://us-assets.i.posthog.com/static/:path*',
-      },
-      {
-        source: '/ingest/:path*',
-        destination: 'https://us.i.posthog.com/:path*',
-      },
-    ]
-  },
 }
 
 export default nextConfig

apps/sim/providers/anthropic/index.ts

@@ -109,9 +109,15 @@ export const anthropicProvider: ProviderConfig = {
             ],
           })
         } else {
+          // Handle content that's already in array format (from transformAttachmentMessages)
+          const content = Array.isArray(msg.content)
+            ? msg.content
+            : msg.content
+              ? [{ type: 'text', text: msg.content }]
+              : []
           messages.push({
             role: msg.role === 'assistant' ? 'assistant' : 'user',
-            content: msg.content ? [{ type: 'text', text: msg.content }] : [],
+            content,
           })
         }
       })

apps/sim/providers/attachment.ts

@@ -0,0 +1,397 @@
+/**
+ * Centralized attachment content transformation for all providers.
+ *
+ * Strategy: Always normalize to base64 first, then create provider-specific formats.
+ * This eliminates URL accessibility issues and simplifies provider handling.
+ */
+
+import { createLogger } from '@sim/logger'
+import { bufferToBase64 } from '@/lib/uploads/utils/file-utils'
+import { downloadFileFromUrl } from '@/lib/uploads/utils/file-utils.server'
+import { supportsVision } from '@/providers/models'
+import type { ProviderId } from '@/providers/types'
+
+const logger = createLogger('AttachmentTransformer')
+
+/**
+ * Generic message type for attachment transformation.
+ */
+interface TransformableMessage {
+  role: string
+  content: string | any[] | null
+  attachment?: AttachmentContent
+  [key: string]: any
+}
+
+/**
+ * Attachment content (files, images, documents)
+ */
+export interface AttachmentContent {
+  sourceType: 'url' | 'base64' | 'file'
+  data: string
+  mimeType?: string
+  fileName?: string
+}
+
+/**
+ * Normalized attachment data (always base64)
+ */
+interface NormalizedAttachment {
+  base64: string
+  mimeType: string
+}
+
+/**
+ * Configuration for attachment transformation
+ */
+interface AttachmentTransformConfig {
+  providerId: ProviderId
+  model: string
+}
+
+/**
+ * Checks if a model supports attachments (vision/multimodal content).
+ */
+export function modelSupportsAttachments(model: string): boolean {
+  return supportsVision(model)
+}
+
+/**
+ * Transforms messages with 'attachment' role into provider-compatible format.
+ */
+export async function transformAttachmentMessages<T extends TransformableMessage>(
+  messages: T[],
+  config: AttachmentTransformConfig
+): Promise<T[]> {
+  const { providerId, model } = config
+  const supportsAttachments = modelSupportsAttachments(model)
+
+  if (!supportsAttachments) {
+    return transformAttachmentsToText(messages) as T[]
+  }
+
+  const result: T[] = []
+
+  for (const msg of messages) {
+    if (msg.role !== 'attachment') {
+      result.push(msg)
+      continue
+    }
+
+    const attachmentContent = await createProviderAttachmentContent(msg, providerId)
+    if (!attachmentContent) {
+      logger.warn('Could not create attachment content for message', { msg })
+      continue
+    }
+
+    // Merge with previous user message or create new one
+    const lastMessage = result[result.length - 1]
+    if (lastMessage && lastMessage.role === 'user') {
+      const existingContent = ensureContentArray(lastMessage, providerId)
+      existingContent.push(attachmentContent)
+      lastMessage.content = existingContent as any
+    } else {
+      result.push({
+        role: 'user',
+        content: [attachmentContent] as any,
+      } as T)
+    }
+  }
+
+  // Ensure all user messages have consistent content format
+  return result.map((msg) => {
+    if (msg.role === 'user' && typeof msg.content === 'string') {
+      return {
+        ...msg,
+        content: [createTextContent(msg.content, providerId)] as any,
+      }
+    }
+    return msg
+  })
+}
+
+/**
+ * Transforms attachment messages to text placeholders for non-vision models
+ */
+function transformAttachmentsToText<T extends TransformableMessage>(messages: T[]): T[] {
+  const result: T[] = []
+
+  for (const msg of messages) {
+    if (msg.role !== 'attachment') {
+      result.push(msg)
+      continue
+    }
+
+    const attachment = msg.attachment
+    const mimeType = attachment?.mimeType || 'unknown type'
+    const fileName = attachment?.fileName || 'file'
+
+    const lastMessage = result[result.length - 1]
+    if (lastMessage && lastMessage.role === 'user') {
+      const currentContent = typeof lastMessage.content === 'string' ? lastMessage.content : ''
+      lastMessage.content = `${currentContent}\n[Attached file: ${fileName} (${mimeType}) - Note: This model does not support file/image inputs]`
+    } else {
+      result.push({
+        role: 'user',
+        content: `[Attached file: ${fileName} (${mimeType}) - Note: This model does not support file/image inputs]`,
+      } as T)
+    }
+  }
+
+  return result
+}
+
+/**
+ * Ensures a user message has content as an array for multimodal support
+ */
+function ensureContentArray(msg: TransformableMessage, providerId: ProviderId): any[] {
+  if (Array.isArray(msg.content)) {
+    return msg.content
+  }
+  if (typeof msg.content === 'string' && msg.content) {
+    return [createTextContent(msg.content, providerId)]
+  }
+  return []
+}
+
+/**
+ * Creates provider-specific text content block
+ */
+export function createTextContent(text: string, providerId: ProviderId): any {
+  switch (providerId) {
+    case 'google':
+    case 'vertex':
+      return { text }
+    default:
+      return { type: 'text', text }
+  }
+}
+
+/**
+ * Normalizes attachment data to base64.
+ * Fetches URLs and converts to base64, extracts base64 from data URLs.
+ */
+async function normalizeToBase64(
+  attachment: AttachmentContent
+): Promise<NormalizedAttachment | null> {
+  const { sourceType, data, mimeType } = attachment
+
+  if (!data || !data.trim()) {
+    logger.warn('Empty attachment data')
+    return null
+  }
+
+  const trimmedData = data.trim()
+
+  // Already base64
+  if (sourceType === 'base64') {
+    // Handle data URL format: data:mime;base64,xxx
+    if (trimmedData.startsWith('data:')) {
+      const match = trimmedData.match(/^data:([^;]+);base64,(.+)$/)
+      if (match) {
+        return { base64: match[2], mimeType: match[1] }
+      }
+    }
+    // Raw base64
+    return { base64: trimmedData, mimeType: mimeType || 'application/octet-stream' }
+  }
+
+  // URL or file path - need to fetch
+  if (sourceType === 'url' || sourceType === 'file') {
+    try {
+      logger.info('Fetching attachment for base64 conversion', {
+        url: trimmedData.substring(0, 50),
+      })
+      const buffer = await downloadFileFromUrl(trimmedData)
+      const base64 = bufferToBase64(buffer)
+      return { base64, mimeType: mimeType || 'application/octet-stream' }
+    } catch (error) {
+      logger.error('Failed to fetch attachment', { error, url: trimmedData.substring(0, 50) })
+      return null
+    }
+  }
+
+  return null
+}
+
+/**
+ * Creates provider-specific attachment content from an attachment message.
+ * First normalizes to base64, then creates the provider format.
+ */
+async function createProviderAttachmentContent(
+  msg: TransformableMessage,
+  providerId: ProviderId
+): Promise<any> {
+  const attachment = msg.attachment
+  if (!attachment) return null
+
+  // Normalize to base64 first
+  const normalized = await normalizeToBase64(attachment)
+  if (!normalized) {
+    return createTextContent('[Failed to load attachment]', providerId)
+  }
+
+  const { base64, mimeType } = normalized
+
+  switch (providerId) {
+    case 'anthropic':
+      return createAnthropicContent(base64, mimeType)
+
+    case 'google':
+    case 'vertex':
+      return createGeminiContent(base64, mimeType)
+
+    case 'mistral':
+      return createMistralContent(base64, mimeType)
+
+    case 'bedrock':
+      return createBedrockContent(base64, mimeType)
+
+    default:
+      // OpenAI format (OpenAI, Azure, xAI, DeepSeek, Cerebras, Groq, OpenRouter, Ollama, vLLM)
+      return createOpenAIContent(base64, mimeType)
+  }
+}
+
+/**
+ * OpenAI-compatible content (images only via base64 data URL)
+ */
+function createOpenAIContent(base64: string, mimeType: string): any {
+  const isImage = mimeType.startsWith('image/')
+  const isAudio = mimeType.startsWith('audio/')
+
+  if (isImage) {
+    return {
+      type: 'image_url',
+      image_url: {
+        url: `data:${mimeType};base64,${base64}`,
+        detail: 'auto',
+      },
+    }
+  }
+
+  if (isAudio) {
+    return {
+      type: 'input_audio',
+      input_audio: {
+        data: base64,
+        format: mimeType === 'audio/wav' ? 'wav' : 'mp3',
+      },
+    }
+  }
+
+  // OpenAI Chat API doesn't support other file types directly
+  // For PDFs/docs, return a text placeholder
+  logger.warn(`OpenAI does not support ${mimeType} attachments in Chat API`)
+  return {
+    type: 'text',
+    text: `[Attached file: ${mimeType} - OpenAI Chat API only supports images and audio]`,
+  }
+}
+
+/**
+ * Anthropic-compatible content (images and PDFs)
+ */
+function createAnthropicContent(base64: string, mimeType: string): any {
+  const isImage = mimeType.startsWith('image/')
+  const isPdf = mimeType === 'application/pdf'
+
+  if (isImage) {
+    return {
+      type: 'image',
+      source: {
+        type: 'base64',
+        media_type: mimeType,
+        data: base64,
+      },
+    }
+  }
+
+  if (isPdf) {
+    return {
+      type: 'document',
+      source: {
+        type: 'base64',
+        media_type: 'application/pdf',
+        data: base64,
+      },
+    }
+  }
+
+  return {
+    type: 'text',
+    text: `[Attached file: ${mimeType} - Anthropic supports images and PDFs only]`,
+  }
+}
+
+/**
+ * Google Gemini-compatible content (inlineData format)
+ */
+function createGeminiContent(base64: string, mimeType: string): any {
+  // Gemini supports a wide range of file types via inlineData
+  return {
+    inlineData: {
+      mimeType,
+      data: base64,
+    },
+  }
+}
+
+/**
+ * Mistral-compatible content (images only, data URL format)
+ */
+function createMistralContent(base64: string, mimeType: string): any {
+  const isImage = mimeType.startsWith('image/')
+
+  if (isImage) {
+    // Mistral uses direct string for image_url, not nested object
+    return {
+      type: 'image_url',
+      image_url: `data:${mimeType};base64,${base64}`,
+    }
+  }
+
+  return {
+    type: 'text',
+    text: `[Attached file: ${mimeType} - Mistral supports images only]`,
+  }
+}
+
+/**
+ * AWS Bedrock-compatible content (images and PDFs)
+ */
+function createBedrockContent(base64: string, mimeType: string): any {
+  const isImage = mimeType.startsWith('image/')
+  const isPdf = mimeType === 'application/pdf'
+
+  // Determine image format from mimeType
+  const getImageFormat = (mime: string): string => {
+    if (mime.includes('jpeg') || mime.includes('jpg')) return 'jpeg'
+    if (mime.includes('png')) return 'png'
+    if (mime.includes('gif')) return 'gif'
+    if (mime.includes('webp')) return 'webp'
+    return 'png'
+  }
+
+  if (isImage) {
+    // Return a marker object that the Bedrock provider will convert to proper format
+    return {
+      type: 'bedrock_image',
+      format: getImageFormat(mimeType),
+      data: base64,
+    }
+  }
+
+  if (isPdf) {
+    return {
+      type: 'bedrock_document',
+      format: 'pdf',
+      data: base64,
+    }
+  }
+
+  return {
+    type: 'text',
+    text: `[Attached file: ${mimeType} - Bedrock supports images and PDFs only]`,
+  }
+}

apps/sim/providers/bedrock/index.ts

@@ -16,6 +16,7 @@ import type { StreamingExecution } from '@/executor/types'
 import { MAX_TOOL_ITERATIONS } from '@/providers'
 import {
   checkForForcedToolUsage,
+  convertToBedrockContentBlocks,
   createReadableStreamFromBedrockStream,
   generateToolUseId,
   getBedrockInferenceProfileId,
@@ -116,9 +117,11 @@ export const bedrockProvider: ProviderConfig = {
           }
         } else {
           const role: ConversationRole = msg.role === 'assistant' ? 'assistant' : 'user'
+          // Handle multimodal content arrays
+          const contentBlocks = convertToBedrockContentBlocks(msg.content || '')
           messages.push({
             role,
-            content: [{ text: msg.content || '' }],
+            content: contentBlocks,
           })
         }
       }

apps/sim/providers/bedrock/utils.ts

@@ -1,9 +1,199 @@
-import type { ConverseStreamOutput } from '@aws-sdk/client-bedrock-runtime'
+import type {
+  ContentBlock,
+  ConverseStreamOutput,
+  ImageFormat,
+} from '@aws-sdk/client-bedrock-runtime'
 import { createLogger } from '@sim/logger'
 import { trackForcedToolUsage } from '@/providers/utils'
 
 const logger = createLogger('BedrockUtils')
 
+/**
+ * Converts message content (string or array) to Bedrock ContentBlock array.
+ * Handles multimodal content including images and documents.
+ */
+export function convertToBedrockContentBlocks(content: string | any[]): ContentBlock[] {
+  // Simple string content
+  if (typeof content === 'string') {
+    return [{ text: content || '' }]
+  }
+
+  // Array content - could be multimodal
+  if (!Array.isArray(content)) {
+    return [{ text: String(content) || '' }]
+  }
+
+  const blocks: ContentBlock[] = []
+
+  for (const item of content) {
+    if (!item) continue
+
+    // Text content
+    if (item.type === 'text' && item.text) {
+      blocks.push({ text: item.text })
+      continue
+    }
+
+    // Gemini-style text (just { text: "..." })
+    if (typeof item.text === 'string' && !item.type) {
+      blocks.push({ text: item.text })
+      continue
+    }
+
+    // Bedrock image content (from agent handler)
+    if (item.type === 'bedrock_image') {
+      const imageBlock = createBedrockImageBlock(item)
+      if (imageBlock) {
+        blocks.push(imageBlock)
+      }
+      continue
+    }
+
+    // Bedrock document content (from agent handler)
+    if (item.type === 'bedrock_document') {
+      const docBlock = createBedrockDocumentBlock(item)
+      if (docBlock) {
+        blocks.push(docBlock)
+      }
+      continue
+    }
+
+    // OpenAI-style image_url (fallback for direct OpenAI format)
+    if (item.type === 'image_url' && item.image_url) {
+      const url = typeof item.image_url === 'string' ? item.image_url : item.image_url?.url
+      if (url) {
+        const imageBlock = createBedrockImageBlockFromUrl(url)
+        if (imageBlock) {
+          blocks.push(imageBlock)
+        }
+      }
+      continue
+    }
+
+    // Unknown type - log warning and skip
+    logger.warn('Unknown content block type in Bedrock conversion:', { type: item.type })
+  }
+
+  // Ensure at least one text block
+  if (blocks.length === 0) {
+    blocks.push({ text: '' })
+  }
+
+  return blocks
+}
+
+/**
+ * Creates a Bedrock image ContentBlock from a bedrock_image item
+ */
+function createBedrockImageBlock(item: {
+  format: string
+  sourceType: string
+  data?: string
+  url?: string
+}): ContentBlock | null {
+  const format = (item.format || 'png') as ImageFormat
+
+  if (item.sourceType === 'base64' && item.data) {
+    // Convert base64 to Uint8Array
+    const bytes = base64ToUint8Array(item.data)
+    return {
+      image: {
+        format,
+        source: { bytes },
+      },
+    }
+  }
+
+  if (item.sourceType === 'url' && item.url) {
+    // For URLs, we need to fetch the image and convert to bytes
+    // This is a limitation - Bedrock doesn't support URL sources directly
+    // The provider layer should handle this, or we log a warning
+    logger.warn('Bedrock does not support image URLs directly. Image will be skipped.', {
+      url: item.url,
+    })
+    // Return a text placeholder
+    return { text: `[Image from URL: ${item.url}]` }
+  }
+
+  return null
+}
+
+/**
+ * Creates a Bedrock document ContentBlock from a bedrock_document item
+ */
+function createBedrockDocumentBlock(item: {
+  format: string
+  sourceType: string
+  data?: string
+  url?: string
+}): ContentBlock | null {
+  if (item.sourceType === 'base64' && item.data) {
+    const bytes = base64ToUint8Array(item.data)
+    return {
+      document: {
+        format: 'pdf',
+        name: 'document',
+        source: { bytes },
+      },
+    }
+  }
+
+  if (item.sourceType === 'url' && item.url) {
+    logger.warn('Bedrock does not support document URLs directly. Document will be skipped.', {
+      url: item.url,
+    })
+    return { text: `[Document from URL: ${item.url}]` }
+  }
+
+  return null
+}
+
+/**
+ * Creates a Bedrock image ContentBlock from a data URL or regular URL
+ */
+function createBedrockImageBlockFromUrl(url: string): ContentBlock | null {
+  // Check if it's a data URL (base64)
+  if (url.startsWith('data:')) {
+    const match = url.match(/^data:image\/(\w+);base64,(.+)$/)
+    if (match) {
+      let format: ImageFormat = match[1] as ImageFormat
+      // Normalize jpg to jpeg
+      if (format === ('jpg' as ImageFormat)) {
+        format = 'jpeg'
+      }
+      const base64Data = match[2]
+      const bytes = base64ToUint8Array(base64Data)
+      return {
+        image: {
+          format,
+          source: { bytes },
+        },
+      }
+    }
+  }
+
+  // Regular URL - Bedrock doesn't support this directly
+  logger.warn('Bedrock does not support image URLs directly. Image will be skipped.', { url })
+  return { text: `[Image from URL: ${url}]` }
+}
+
+/**
+ * Converts a base64 string to Uint8Array
+ */
+function base64ToUint8Array(base64: string): Uint8Array {
+  // Handle browser and Node.js environments
+  if (typeof Buffer !== 'undefined') {
+    return Buffer.from(base64, 'base64')
+  }
+  // Browser fallback
+  const binaryString = atob(base64)
+  const bytes = new Uint8Array(binaryString.length)
+  for (let i = 0; i < binaryString.length; i++) {
+    bytes[i] = binaryString.charCodeAt(i)
+  }
+  return bytes
+}
+
 export interface BedrockStreamUsage {
   inputTokens: number
   outputTokens: number

apps/sim/providers/google/utils.ts

@@ -72,6 +72,75 @@ export function cleanSchemaForGemini(schema: SchemaUnion): SchemaUnion {
   return cleanedSchema
 }
 
+/**
+ * Converts an array of content items to Gemini-compatible Part array.
+ * Handles various formats from the attachment transformer.
+ */
+function convertContentArrayToGeminiParts(contentArray: any[]): Part[] {
+  const parts: Part[] = []
+
+  for (const item of contentArray) {
+    if (!item) continue
+
+    // Gemini-native text format: { text: "..." }
+    if (typeof item.text === 'string') {
+      parts.push({ text: item.text })
+      continue
+    }
+
+    // OpenAI-style text: { type: 'text', text: '...' }
+    if (item.type === 'text' && typeof item.text === 'string') {
+      parts.push({ text: item.text })
+      continue
+    }
+
+    // Gemini-native inlineData format (from attachment transformer)
+    if (item.inlineData) {
+      parts.push({ inlineData: item.inlineData })
+      continue
+    }
+
+    // Gemini-native fileData format (from attachment transformer)
+    if (item.fileData) {
+      parts.push({ fileData: item.fileData })
+      continue
+    }
+
+    // OpenAI-style image_url - convert to Gemini format
+    if (item.type === 'image_url' && item.image_url) {
+      const url = typeof item.image_url === 'string' ? item.image_url : item.image_url?.url
+      if (url) {
+        // Check if it's a data URL (base64)
+        if (url.startsWith('data:')) {
+          const match = url.match(/^data:([^;]+);base64,(.+)$/)
+          if (match) {
+            parts.push({
+              inlineData: {
+                mimeType: match[1],
+                data: match[2],
+              },
+            })
+          }
+        } else {
+          // External URL
+          parts.push({
+            fileData: {
+              mimeType: 'image/jpeg', // Default, Gemini will detect actual type
+              fileUri: url,
+            },
+          })
+        }
+      }
+      continue
+    }
+
+    // Unknown type - log warning
+    logger.warn('Unknown content item type in Gemini conversion:', { type: item.type })
+  }
+
+  return parts
+}
+
 /**
  * Extracts text content from a Gemini response candidate.
  * Filters out thought parts (model reasoning) from the output.
@@ -180,7 +249,13 @@ export function convertToGeminiFormat(request: ProviderRequest): {
       } else if (message.role === 'user' || message.role === 'assistant') {
         const geminiRole = message.role === 'user' ? 'user' : 'model'
 
-        if (message.content) {
+        // Handle multimodal content (arrays with text/image/file parts)
+        if (Array.isArray(message.content)) {
+          const parts: Part[] = convertContentArrayToGeminiParts(message.content)
+          if (parts.length > 0) {
+            contents.push({ role: geminiRole, parts })
+          }
+        } else if (message.content) {
           contents.push({ role: geminiRole, parts: [{ text: message.content }] })
         }
 

apps/sim/providers/models.ts

@@ -34,6 +34,8 @@ export interface ModelCapabilities {
   toolUsageControl?: boolean
   computerUse?: boolean
   nativeStructuredOutputs?: boolean
+  /** Whether the model supports vision/multimodal inputs (images, audio, video, PDFs) */
+  vision?: boolean
   maxOutputTokens?: {
     /** Maximum tokens for streaming requests */
     max: number
@@ -120,6 +122,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -132,6 +135,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-12-11',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'minimal', 'low', 'medium', 'high', 'xhigh'],
           },
@@ -150,6 +154,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-11-14',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'low', 'medium', 'high'],
           },
@@ -222,6 +227,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-08-07',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['minimal', 'low', 'medium', 'high'],
           },
@@ -240,6 +246,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-08-07',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['minimal', 'low', 'medium', 'high'],
           },
@@ -258,6 +265,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-08-07',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['minimal', 'low', 'medium', 'high'],
           },
@@ -287,6 +295,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-06-17',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['low', 'medium', 'high'],
           },
@@ -302,6 +311,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-06-17',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['low', 'medium', 'high'],
           },
@@ -317,6 +327,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-06-17',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['low', 'medium', 'high'],
           },
@@ -333,6 +344,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -346,6 +358,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -359,6 +372,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -385,6 +399,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 2 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -397,6 +412,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-12-11',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'minimal', 'low', 'medium', 'high', 'xhigh'],
           },
@@ -415,6 +431,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-11-14',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'low', 'medium', 'high'],
           },
@@ -433,6 +450,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-11-14',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'low', 'medium', 'high'],
           },
@@ -451,6 +469,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-11-14',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'low', 'medium', 'high'],
           },
@@ -469,6 +488,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-11-14',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['none', 'medium', 'high'],
           },
@@ -487,6 +507,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-08-07',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['minimal', 'low', 'medium', 'high'],
           },
@@ -505,6 +526,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-08-07',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['minimal', 'low', 'medium', 'high'],
           },
@@ -523,6 +545,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-08-07',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['minimal', 'low', 'medium', 'high'],
           },
@@ -552,6 +575,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-06-15',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['low', 'medium', 'high'],
           },
@@ -567,6 +591,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           updatedAt: '2025-06-15',
         },
         capabilities: {
+          vision: true,
           reasoningEffort: {
             values: ['low', 'medium', 'high'],
           },
@@ -581,7 +606,9 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           output: 8.0,
           updatedAt: '2025-06-15',
         },
-        capabilities: {},
+        capabilities: {
+          vision: true,
+        },
         contextWindow: 1000000,
       },
       {
@@ -620,6 +647,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -635,6 +663,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -649,6 +678,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         capabilities: {
           temperature: { min: 0, max: 1 },
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -664,6 +694,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -679,6 +710,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -693,6 +725,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         capabilities: {
           temperature: { min: 0, max: 1 },
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -708,6 +741,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           computerUse: true,
           maxOutputTokens: { max: 8192, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -723,6 +757,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           computerUse: true,
           maxOutputTokens: { max: 8192, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -736,6 +771,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     modelPatterns: [/^gemini/],
     capabilities: {
       toolUsageControl: true,
+      vision: true,
     },
     icon: GeminiIcon,
     models: [
@@ -847,6 +883,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     icon: VertexIcon,
     capabilities: {
       toolUsageControl: true,
+      vision: true,
     },
     models: [
       {
@@ -1005,6 +1042,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
     icon: xAIIcon,
     capabilities: {
       toolUsageControl: true,
+      vision: true,
     },
     models: [
       {
@@ -1277,7 +1315,9 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           output: 0.34,
           updatedAt: '2026-01-27',
         },
-        capabilities: {},
+        capabilities: {
+          vision: true,
+        },
         contextWindow: 131072,
       },
       {
@@ -1287,7 +1327,9 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           output: 0.6,
           updatedAt: '2026-01-27',
         },
-        capabilities: {},
+        capabilities: {
+          vision: true,
+        },
         contextWindow: 131072,
       },
       {
@@ -1369,6 +1411,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1381,6 +1424,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1453,6 +1497,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1465,6 +1510,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1489,6 +1535,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1501,6 +1548,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1549,6 +1597,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1561,6 +1610,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1585,6 +1635,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1597,6 +1648,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1609,6 +1661,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1621,6 +1674,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1645,6 +1699,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1657,6 +1712,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 256000,
       },
@@ -1710,6 +1766,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -1724,6 +1781,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -1738,6 +1796,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -1752,6 +1811,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
           temperature: { min: 0, max: 1 },
           nativeStructuredOutputs: true,
           maxOutputTokens: { max: 64000, default: 8192 },
+          vision: true,
         },
         contextWindow: 200000,
       },
@@ -1764,6 +1824,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -1776,6 +1837,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -1788,6 +1850,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -1800,6 +1863,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 300000,
       },
@@ -1812,6 +1876,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 300000,
       },
@@ -1836,6 +1901,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 1000000,
       },
@@ -1848,6 +1914,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 3500000,
       },
@@ -1872,6 +1939,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1884,6 +1952,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1956,6 +2025,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -1992,6 +2062,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -2016,6 +2087,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -2028,6 +2100,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -2040,6 +2113,7 @@ export const PROVIDER_DEFINITIONS: Record<string, ProviderDefinition> = {
         },
         capabilities: {
           temperature: { min: 0, max: 1 },
+          vision: true,
         },
         contextWindow: 128000,
       },
@@ -2211,6 +2285,32 @@ export function getMaxTemperature(modelId: string): number | undefined {
   return capabilities?.temperature?.max
 }
 
+/**
+ * Checks if a model supports vision/multimodal inputs (images, audio, video, PDFs)
+ */
+export function supportsVision(modelId: string): boolean {
+  const capabilities = getModelCapabilities(modelId)
+  return !!capabilities?.vision
+}
+
+/**
+ * Returns a list of all vision-capable models
+ */
+export function getVisionModels(): string[] {
+  const models: string[] = []
+  for (const provider of Object.values(PROVIDER_DEFINITIONS)) {
+    // Check if the provider has vision capability at the provider level
+    const providerHasVision = provider.capabilities?.vision
+    for (const model of provider.models) {
+      // Model has vision if either the model or provider has vision capability
+      if (model.capabilities.vision || providerHasVision) {
+        models.push(model.id)
+      }
+    }
+  }
+  return models
+}
+
 export function supportsToolUsageControl(providerId: string): boolean {
   return getProvidersWithToolUsageControl().includes(providerId)
 }

apps/sim/providers/types.ts

@@ -111,9 +111,25 @@ export interface ProviderToolConfig {
   usageControl?: ToolUsageControl
 }
 
+/**
+ * Attachment content (files, images, documents)
+ */
+export interface AttachmentContent {
+  /** Source type: how the data was provided */
+  sourceType: 'url' | 'base64' | 'file'
+  /** The URL or base64 data */
+  data: string
+  /** MIME type (e.g., 'image/png', 'application/pdf', 'audio/mp3') */
+  mimeType?: string
+  /** Optional filename for file uploads */
+  fileName?: string
+}
+
 export interface Message {
-  role: 'system' | 'user' | 'assistant' | 'function' | 'tool'
+  role: 'system' | 'user' | 'assistant' | 'function' | 'tool' | 'attachment'
   content: string | null
+  /** Attachment content for 'attachment' role messages */
+  attachment?: AttachmentContent
   name?: string
   function_call?: {
     name: string

apps/sim/providers/utils.ts

@@ -23,9 +23,11 @@ import {
   getReasoningEffortValuesForModel as getReasoningEffortValuesForModelFromDefinitions,
   getThinkingLevelsForModel as getThinkingLevelsForModelFromDefinitions,
   getVerbosityValuesForModel as getVerbosityValuesForModelFromDefinitions,
+  getVisionModels,
   PROVIDER_DEFINITIONS,
   supportsTemperature as supportsTemperatureFromDefinitions,
   supportsToolUsageControl as supportsToolUsageControlFromDefinitions,
+  supportsVision,
   updateOllamaModels as updateOllamaModelsInDefinitions,
 } from '@/providers/models'
 import type { ProviderId, ProviderToolConfig } from '@/providers/types'
@@ -1152,3 +1154,6 @@ export function checkForForcedToolUsageOpenAI(
 
   return { hasUsedForcedTool, usedForcedTools: updatedUsedForcedTools }
 }
+
+// Re-export vision capability functions
+export { supportsVision, getVisionModels }

apps/sim/proxy.ts

@@ -134,6 +134,24 @@ function handleSecurityFiltering(request: NextRequest): NextResponse | null {
 export async function proxy(request: NextRequest) {
   const url = request.nextUrl
 
+  if (url.pathname.startsWith('/ingest/')) {
+    const hostname = url.pathname.startsWith('/ingest/static/')
+      ? 'us-assets.i.posthog.com'
+      : 'us.i.posthog.com'
+
+    const targetPath = url.pathname.replace(/^\/ingest/, '')
+    const targetUrl = `https://${hostname}${targetPath}${url.search}`
+
+    return NextResponse.rewrite(new URL(targetUrl), {
+      request: {
+        headers: new Headers({
+          ...Object.fromEntries(request.headers),
+          host: hostname,
+        }),
+      },
+    })
+  }
+
   const sessionCookie = getSessionCookie(request)
   const hasActiveSession = isAuthDisabled || !!sessionCookie
 
@@ -195,6 +213,7 @@ export async function proxy(request: NextRequest) {
 
 export const config = {
   matcher: [
+    '/ingest/:path*', // PostHog proxy for session recording
     '/', // Root path for self-hosted redirect logic
     '/terms', // Whitelabel terms redirect
     '/privacy', // Whitelabel privacy redirect

apps/sim/stores/panel/copilot/store.ts

@@ -54,6 +54,7 @@ import { TestClientTool } from '@/lib/copilot/tools/client/other/test'
 import { TourClientTool } from '@/lib/copilot/tools/client/other/tour'
 import { WorkflowClientTool } from '@/lib/copilot/tools/client/other/workflow'
 import { createExecutionContext, getTool } from '@/lib/copilot/tools/client/registry'
+import { COPILOT_SERVER_ORCHESTRATED } from '@/lib/copilot/orchestrator/config'
 import { GetCredentialsClientTool } from '@/lib/copilot/tools/client/user/get-credentials'
 import { SetEnvironmentVariablesClientTool } from '@/lib/copilot/tools/client/user/set-environment-variables'
 import { CheckDeploymentStatusClientTool } from '@/lib/copilot/tools/client/workflow/check-deployment-status'
@@ -1198,6 +1199,18 @@ const sseHandlers: Record<string, SSEHandler> = {
             }
           } catch {}
         }
+
+        if (COPILOT_SERVER_ORCHESTRATED && current.name === 'edit_workflow') {
+          try {
+            const resultPayload =
+              data?.result || data?.data?.result || data?.data?.data || data?.data || {}
+            const workflowState = resultPayload?.workflowState
+            if (workflowState) {
+              const diffStore = useWorkflowDiffStore.getState()
+              void diffStore.setProposedChanges(workflowState)
+            }
+          } catch {}
+        }
       }
 
       // Update inline content block state
@@ -1362,6 +1375,10 @@ const sseHandlers: Record<string, SSEHandler> = {
       return
     }
 
+    if (COPILOT_SERVER_ORCHESTRATED) {
+      return
+    }
+
     // Prefer interface-based registry to determine interrupt and execute
     try {
       const def = name ? getTool(name) : undefined
@@ -3820,6 +3837,9 @@ export const useCopilotStore = create<CopilotStore>()(
     setEnabledModels: (models) => set({ enabledModels: models }),
 
     executeIntegrationTool: async (toolCallId: string) => {
+      if (COPILOT_SERVER_ORCHESTRATED) {
+        return
+      }
       const { toolCallsById, workflowId } = get()
       const toolCall = toolCallsById[toolCallId]
       if (!toolCall || !workflowId) return

docs/COPILOT_SERVER_REFACTOR.md

@@ -0,0 +1,927 @@
+# Copilot Server-Side Refactor Plan
+
+> **Goal**: Move copilot orchestration logic from the browser (React/Zustand) to the Next.js server, enabling both headless API access and a simplified interactive client.
+
+## Table of Contents
+
+1. [Executive Summary](#executive-summary)
+2. [Current Architecture](#current-architecture)
+3. [Target Architecture](#target-architecture)
+4. [Scope & Boundaries](#scope--boundaries)
+5. [Module Design](#module-design)
+6. [Implementation Plan](#implementation-plan)
+7. [API Contracts](#api-contracts)
+8. [Migration Strategy](#migration-strategy)
+9. [Testing Strategy](#testing-strategy)
+10. [Risks & Mitigations](#risks--mitigations)
+11. [File Inventory](#file-inventory)
+
+---
+
+## Executive Summary
+
+### Problem
+
+The current copilot implementation in Sim has all orchestration logic in the browser:
+- SSE stream parsing happens in the React client
+- Tool execution is triggered from the browser
+- OAuth tokens are sent to the client
+- No headless/API access is possible
+- The Zustand store is ~4,200 lines of complex async logic
+
+### Solution
+
+Move orchestration to the Next.js server:
+- Server parses SSE from copilot backend
+- Server executes tools directly (no HTTP round-trips)
+- Server forwards events to client (if attached)
+- Headless API returns JSON response
+- Client store becomes a thin UI layer (~600 lines)
+
+### Benefits
+
+| Aspect | Before | After |
+|--------|--------|-------|
+| Security | OAuth tokens in browser | Tokens stay server-side |
+| Headless access | Not possible | Full API support |
+| Store complexity | ~4,200 lines | ~600 lines |
+| Tool execution | Browser-initiated | Server-side |
+| Testing | Complex async | Simple state |
+| Bundle size | Large (tool classes) | Minimal |
+
+---
+
+## Current Architecture
+
+```
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                              BROWSER (React)                                 │
+├─────────────────────────────────────────────────────────────────────────────┤
+│                                                                             │
+│  ┌─────────────────────────────────────────────────────────────────────────┐│
+│  │                    Copilot Store (4,200 lines)                          ││
+│  │                                                                         ││
+│  │  • SSE stream parsing (parseSSEStream)                                  ││
+│  │  • Event handlers (sseHandlers, subAgentSSEHandlers)                    ││
+│  │  • Tool execution logic                                                 ││
+│  │  • Client tool instantiation                                            ││
+│  │  • Content block processing                                             ││
+│  │  • State management                                                     ││
+│  │  • UI state                                                             ││
+│  └─────────────────────────────────────────────────────────────────────────┘│
+│         │                                                                   │
+│         │ HTTP calls for tool execution                                     │
+│         ▼                                                                   │
+└─────────────────────────────────────────────────────────────────────────────┘
+          │
+          ▼
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                           NEXT.JS SERVER                                     │
+├─────────────────────────────────────────────────────────────────────────────┤
+│                                                                             │
+│  /api/copilot/chat              - Proxy to copilot backend (pass-through)   │
+│  /api/copilot/execute-tool      - Execute integration tools                 │
+│  /api/copilot/confirm           - Update Redis with tool status             │
+│  /api/copilot/tools/mark-complete - Notify copilot backend                  │
+│  /api/copilot/execute-copilot-server-tool - Execute server tools            │
+│                                                                             │
+└─────────────────────────────────────────────────────────────────────────────┘
+          │
+          ▼
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                        COPILOT BACKEND (Go)                                  │
+│                         copilot.sim.ai                                       │
+├─────────────────────────────────────────────────────────────────────────────┤
+│                                                                             │
+│  • LLM orchestration                                                        │
+│  • Subagent system (plan, edit, debug, etc.)                                │
+│  • Tool definitions                                                         │
+│  • Conversation management                                                  │
+│  • SSE streaming                                                            │
+│                                                                             │
+└─────────────────────────────────────────────────────────────────────────────┘
+```
+
+### Current Flow (Interactive)
+
+1. User sends message in UI
+2. Store calls `/api/copilot/chat`
+3. Chat route proxies to copilot backend, streams SSE back
+4. **Store parses SSE in browser**
+5. On `tool_call` event:
+   - Store decides if tool needs confirmation
+   - Store calls `/api/copilot/execute-tool` or `/api/copilot/execute-copilot-server-tool`
+   - Store calls `/api/copilot/tools/mark-complete`
+6. Store updates UI state
+
+### Problems with Current Flow
+
+1. **No headless access**: Must have browser client
+2. **Security**: OAuth tokens sent to browser for tool execution
+3. **Complexity**: All orchestration logic in Zustand store
+4. **Performance**: Multiple HTTP round-trips from browser
+5. **Reliability**: Browser can disconnect mid-operation
+6. **Testing**: Hard to test async browser logic
+
+---
+
+## Target Architecture
+
+```
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                              BROWSER (React)                                 │
+├─────────────────────────────────────────────────────────────────────────────┤
+│                                                                             │
+│  ┌─────────────────────────────────────────────────────────────────────────┐│
+│  │                    Copilot Store (~600 lines)                           ││
+│  │                                                                         ││
+│  │  • UI state (messages, toolCalls display)                               ││
+│  │  • Event listener (receive server events)                               ││
+│  │  • User actions (send message, confirm/reject)                          ││
+│  │  • Simple API calls                                                     ││
+│  └─────────────────────────────────────────────────────────────────────────┘│
+│         │                                                                   │
+│         │ SSE events from server                                            │
+│         │                                                                   │
+└─────────────────────────────────────────────────────────────────────────────┘
+          ▲
+          │ (Optional - headless mode has no client)
+          │
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                           NEXT.JS SERVER                                     │
+├─────────────────────────────────────────────────────────────────────────────┤
+│                                                                             │
+│  ┌─────────────────────────────────────────────────────────────────────────┐│
+│  │                    Orchestrator Module (NEW)                            ││
+│  │                    lib/copilot/orchestrator/                            ││
+│  │                                                                         ││
+│  │  • SSE stream parsing                                                   ││
+│  │  • Event handlers                                                       ││
+│  │  • Tool execution (direct function calls)                               ││
+│  │  • Response building                                                    ││
+│  │  • Event forwarding (to client if attached)                             ││
+│  └─────────────────────────────────────────────────────────────────────────┘│
+│         │                                                                   │
+│  ┌──────┴──────┐                                                            │
+│  │             │                                                            │
+│  ▼             ▼                                                            │
+│  /api/copilot/chat        /api/v1/copilot/chat                              │
+│  (Interactive)            (Headless)                                        │
+│  - Session auth           - API key auth                                    │
+│  - SSE to client          - JSON response                                   │
+│                                                                             │
+└─────────────────────────────────────────────────────────────────────────────┘
+          │
+          │ (Single external HTTP call)
+          ▼
+┌─────────────────────────────────────────────────────────────────────────────┐
+│                        COPILOT BACKEND (Go)                                  │
+│                    (UNCHANGED - no modifications)                            │
+└─────────────────────────────────────────────────────────────────────────────┘
+```
+
+### Target Flow (Headless)
+
+1. External client calls `POST /api/v1/copilot/chat` with API key
+2. Orchestrator calls copilot backend
+3. **Server parses SSE stream**
+4. **Server executes tools directly** (no HTTP)
+5. Server notifies copilot backend (mark-complete)
+6. Server returns JSON response
+
+### Target Flow (Interactive)
+
+1. User sends message in UI
+2. Store calls `/api/copilot/chat`
+3. **Server orchestrates everything**
+4. Server forwards events to client via SSE
+5. Client just updates UI from events
+6. Server returns when complete
+
+---
+
+## Scope & Boundaries
+
+### In Scope
+
+| Item | Description |
+|------|-------------|
+| Orchestrator module | New module in `lib/copilot/orchestrator/` |
+| Headless API route | New route `POST /api/v1/copilot/chat` |
+| SSE parsing | Move from store to server |
+| Tool execution | Direct function calls on server |
+| Event forwarding | SSE to client (interactive mode) |
+| Store simplification | Reduce to UI-only logic |
+
+### Out of Scope
+
+| Item | Reason |
+|------|--------|
+| Copilot backend (Go) | Separate repo, working correctly |
+| Tool definitions | Already work, just called differently |
+| LLM providers | Handled by copilot backend |
+| Subagent system | Handled by copilot backend |
+
+### Boundaries
+
+```
+                    ┌─────────────────────────────────────┐
+                    │         MODIFICATION ZONE           │
+                    │                                     │
+   ┌────────────────┼─────────────────────────────────────┼────────────────┐
+   │                │                                     │                │
+   │  UNCHANGED     │   apps/sim/                         │   UNCHANGED    │
+   │                │   ├── lib/copilot/orchestrator/     │                │
+   │  copilot/      │   │   └── (NEW)                     │   apps/sim/    │
+   │  (Go backend)  │   ├── app/api/v1/copilot/           │   tools/       │
+   │                │   │   └── (NEW)                     │   (definitions)│
+   │                │   ├── app/api/copilot/chat/         │                │
+   │                │   │   └── (MODIFIED)                │                │
+   │                │   └── stores/panel/copilot/         │                │
+   │                │       └── (SIMPLIFIED)              │                │
+   │                │                                     │                │
+   └────────────────┼─────────────────────────────────────┼────────────────┘
+                    │                                     │
+                    └─────────────────────────────────────┘
+```
+
+---
+
+## Module Design
+
+### Directory Structure
+
+```
+apps/sim/lib/copilot/orchestrator/
+├── index.ts              # Main orchestrator function
+├── types.ts              # Type definitions
+├── sse-parser.ts         # Parse SSE stream from copilot backend
+├── sse-handlers.ts       # Handle each SSE event type
+├── tool-executor.ts      # Execute tools directly (no HTTP)
+├── persistence.ts        # Database and Redis operations
+└── response-builder.ts   # Build final response
+```
+
+### Module Responsibilities
+
+#### `types.ts`
+
+Defines all types used by the orchestrator:
+
+```typescript
+// SSE Events
+interface SSEEvent { type, data, subagent?, toolCallId?, toolName? }
+type SSEEventType = 'content' | 'tool_call' | 'tool_result' | 'done' | ...
+
+// Tool State
+interface ToolCallState { id, name, status, params?, result?, error? }
+type ToolCallStatus = 'pending' | 'executing' | 'success' | 'error' | 'skipped'
+
+// Streaming Context (internal state during orchestration)
+interface StreamingContext { 
+  chatId?, conversationId?, messageId
+  accumulatedContent, contentBlocks
+  toolCalls: Map<string, ToolCallState>
+  streamComplete, errors[]
+}
+
+// Orchestrator API
+interface OrchestratorRequest { message, workflowId, userId, chatId?, mode?, ... }
+interface OrchestratorOptions { autoExecuteTools?, onEvent?, timeout?, ... }
+interface OrchestratorResult { success, content, toolCalls[], chatId?, error? }
+
+// Execution Context (passed to tool executors)
+interface ExecutionContext { userId, workflowId, workspaceId?, decryptedEnvVars? }
+```
+
+#### `sse-parser.ts`
+
+Parses SSE stream into typed events:
+
+```typescript
+async function* parseSSEStream(
+  reader: ReadableStreamDefaultReader,
+  decoder: TextDecoder,
+  abortSignal?: AbortSignal
+): AsyncGenerator<SSEEvent>
+```
+
+- Handles buffering for partial lines
+- Parses JSON from `data:` lines
+- Yields typed `SSEEvent` objects
+- Supports abort signal
+
+#### `sse-handlers.ts`
+
+Handles each SSE event type:
+
+```typescript
+const sseHandlers: Record<SSEEventType, SSEHandler> = {
+  content: (event, context) => { /* append to accumulated content */ },
+  tool_call: async (event, context, execContext, options) => { 
+    /* track tool, execute if autoExecuteTools */ 
+  },
+  tool_result: (event, context) => { /* update tool status */ },
+  tool_generating: (event, context) => { /* create pending tool */ },
+  reasoning: (event, context) => { /* handle thinking blocks */ },
+  done: (event, context) => { /* mark stream complete */ },
+  error: (event, context) => { /* record error */ },
+  // ... etc
+}
+
+const subAgentHandlers: Record<SSEEventType, SSEHandler> = {
+  // Handlers for events within subagent context
+}
+```
+
+#### `tool-executor.ts`
+
+Executes tools directly without HTTP:
+
+```typescript
+// Main entry point
+async function executeToolServerSide(
+  toolCall: ToolCallState,
+  context: ExecutionContext
+): Promise<ToolCallResult>
+
+// Server tools (edit_workflow, search_documentation, etc.)
+async function executeServerToolDirect(
+  toolName: string,
+  params: Record<string, any>,
+  context: ExecutionContext
+): Promise<ToolCallResult>
+
+// Integration tools (slack_send, gmail_read, etc.)
+async function executeIntegrationToolDirect(
+  toolCallId: string,
+  toolName: string,
+  toolConfig: ToolConfig,
+  params: Record<string, any>,
+  context: ExecutionContext
+): Promise<ToolCallResult>
+
+// Notify copilot backend (external HTTP - required)
+async function markToolComplete(
+  toolCallId: string,
+  toolName: string,
+  status: number,
+  message?: any,
+  data?: any
+): Promise<boolean>
+
+// Prepare cached context for tool execution
+async function prepareExecutionContext(
+  userId: string,
+  workflowId: string
+): Promise<ExecutionContext>
+```
+
+**Key principle**: Internal tool execution uses direct function calls. Only `markToolComplete` makes HTTP call (to copilot backend - external).
+
+#### `persistence.ts`
+
+Database and Redis operations:
+
+```typescript
+// Chat persistence
+async function createChat(params): Promise<{ id: string }>
+async function loadChat(chatId, userId): Promise<Chat | null>
+async function saveMessages(chatId, messages, options?): Promise<void>
+async function updateChatConversationId(chatId, conversationId): Promise<void>
+
+// Tool confirmation (Redis)
+async function setToolConfirmation(toolCallId, status, message?): Promise<boolean>
+async function getToolConfirmation(toolCallId): Promise<Confirmation | null>
+```
+
+#### `index.ts`
+
+Main orchestrator function:
+
+```typescript
+async function orchestrateCopilotRequest(
+  request: OrchestratorRequest,
+  options: OrchestratorOptions = {}
+): Promise<OrchestratorResult> {
+  
+  // 1. Prepare execution context (cache env vars, etc.)
+  const execContext = await prepareExecutionContext(userId, workflowId)
+  
+  // 2. Handle chat creation/loading
+  let chatId = await resolveChat(request)
+  
+  // 3. Build request payload for copilot backend
+  const payload = buildCopilotPayload(request)
+  
+  // 4. Call copilot backend
+  const response = await fetch(COPILOT_URL, { body: JSON.stringify(payload) })
+  
+  // 5. Create streaming context
+  const context = createStreamingContext(chatId)
+  
+  // 6. Parse and handle SSE stream
+  for await (const event of parseSSEStream(response.body)) {
+    // Forward to client if attached
+    options.onEvent?.(event)
+    
+    // Handle event
+    const handler = getHandler(event)
+    await handler(event, context, execContext, options)
+    
+    if (context.streamComplete) break
+  }
+  
+  // 7. Persist to database
+  await persistChat(chatId, context)
+  
+  // 8. Build and return result
+  return buildResult(context)
+}
+```
+
+---
+
+## Implementation Plan
+
+### Phase 1: Create Orchestrator Module (3-4 days)
+
+**Goal**: Build the orchestrator module that can run independently.
+
+#### Tasks
+
+1. **Create `types.ts`** (~200 lines)
+   - [ ] Define SSE event types
+   - [ ] Define tool call state types
+   - [ ] Define streaming context type
+   - [ ] Define orchestrator request/response types
+   - [ ] Define execution context type
+
+2. **Create `sse-parser.ts`** (~80 lines)
+   - [ ] Extract parsing logic from store.ts
+   - [ ] Add abort signal support
+   - [ ] Add error handling
+
+3. **Create `persistence.ts`** (~120 lines)
+   - [ ] Extract DB operations from chat route
+   - [ ] Extract Redis operations from confirm route
+   - [ ] Add chat creation/loading
+   - [ ] Add message saving
+
+4. **Create `tool-executor.ts`** (~300 lines)
+   - [ ] Create `executeToolServerSide()` main entry
+   - [ ] Create `executeServerToolDirect()` for server tools
+   - [ ] Create `executeIntegrationToolDirect()` for integration tools
+   - [ ] Create `markToolComplete()` for copilot backend notification
+   - [ ] Create `prepareExecutionContext()` for caching
+   - [ ] Handle OAuth token resolution
+   - [ ] Handle env var resolution
+
+5. **Create `sse-handlers.ts`** (~350 lines)
+   - [ ] Extract handlers from store.ts
+   - [ ] Adapt for server-side context
+   - [ ] Add tool execution integration
+   - [ ] Add subagent handlers
+
+6. **Create `index.ts`** (~250 lines)
+   - [ ] Create `orchestrateCopilotRequest()` main function
+   - [ ] Wire together all modules
+   - [ ] Add timeout handling
+   - [ ] Add abort signal support
+   - [ ] Add event forwarding
+
+#### Deliverables
+
+- Complete `lib/copilot/orchestrator/` module
+- Unit tests for each component
+- Integration test for full orchestration
+
+### Phase 2: Create Headless API Route (1 day)
+
+**Goal**: Create API endpoint for headless copilot access.
+
+#### Tasks
+
+1. **Create route** `app/api/v1/copilot/chat/route.ts` (~100 lines)
+   - [ ] Add API key authentication
+   - [ ] Parse and validate request
+   - [ ] Call orchestrator
+   - [ ] Return JSON response
+
+2. **Add to API documentation**
+   - [ ] Document request format
+   - [ ] Document response format
+   - [ ] Document error codes
+
+#### Deliverables
+
+- Working `POST /api/v1/copilot/chat` endpoint
+- API documentation
+- E2E test
+
+### Phase 3: Wire Interactive Route (2 days)
+
+**Goal**: Use orchestrator for existing interactive flow.
+
+#### Tasks
+
+1. **Modify `/api/copilot/chat/route.ts`**
+   - [ ] Add feature flag for new vs old flow
+   - [ ] Call orchestrator with `onEvent` callback
+   - [ ] Forward events to client via SSE
+   - [ ] Maintain backward compatibility
+
+2. **Test both flows**
+   - [ ] Verify interactive works with new orchestrator
+   - [ ] Verify old flow still works (feature flag off)
+
+#### Deliverables
+
+- Interactive route using orchestrator
+- Feature flag for gradual rollout
+- No breaking changes
+
+### Phase 4: Simplify Client Store (2-3 days)
+
+**Goal**: Remove orchestration logic from client, keep UI-only.
+
+#### Tasks
+
+1. **Create simplified store** (new file or gradual refactor)
+   - [ ] Keep: UI state, messages, tool display
+   - [ ] Keep: Simple API calls
+   - [ ] Keep: Event listener
+   - [ ] Remove: SSE parsing
+   - [ ] Remove: Tool execution logic
+   - [ ] Remove: Client tool instantiators
+
+2. **Update components**
+   - [ ] Update components to use simplified store
+   - [ ] Remove tool execution from UI components
+   - [ ] Simplify tool display components
+
+3. **Remove dead code**
+   - [ ] Remove unused imports
+   - [ ] Remove unused helper functions
+   - [ ] Remove client tool classes (if no longer needed)
+
+#### Deliverables
+
+- Simplified store (~600 lines)
+- Updated components
+- Reduced bundle size
+
+### Phase 5: Testing & Polish (2-3 days)
+
+#### Tasks
+
+1. **E2E testing**
+   - [ ] Test headless API with various prompts
+   - [ ] Test interactive with various prompts
+   - [ ] Test tool execution scenarios
+   - [ ] Test error handling
+   - [ ] Test abort/timeout scenarios
+
+2. **Performance testing**
+   - [ ] Compare latency (old vs new)
+   - [ ] Check memory usage
+   - [ ] Check for connection issues
+
+3. **Documentation**
+   - [ ] Update developer docs
+   - [ ] Add architecture diagram
+   - [ ] Document new API
+
+#### Deliverables
+
+- Comprehensive test suite
+- Performance benchmarks
+- Complete documentation
+
+---
+
+## API Contracts
+
+### Headless API
+
+#### Request
+
+```http
+POST /api/v1/copilot/chat
+Content-Type: application/json
+X-API-Key: sim_xxx
+
+{
+  "message": "Create a Slack notification workflow",
+  "workflowId": "wf_abc123",
+  "chatId": "chat_xyz",           // Optional: continue existing chat
+  "mode": "agent",                // Optional: "agent" | "ask" | "plan"
+  "model": "claude-4-sonnet",     // Optional
+  "autoExecuteTools": true,       // Optional: default true
+  "timeout": 300000               // Optional: default 5 minutes
+}
+```
+
+#### Response (Success)
+
+```json
+{
+  "success": true,
+  "content": "I've created a Slack notification workflow that...",
+  "toolCalls": [
+    {
+      "id": "tc_001",
+      "name": "search_patterns",
+      "status": "success",
+      "params": { "query": "slack notification" },
+      "result": { "patterns": [...] },
+      "durationMs": 234
+    },
+    {
+      "id": "tc_002",
+      "name": "edit_workflow",
+      "status": "success",
+      "params": { "operations": [...] },
+      "result": { "blocksAdded": 3 },
+      "durationMs": 1523
+    }
+  ],
+  "chatId": "chat_xyz",
+  "conversationId": "conv_123"
+}
+```
+
+#### Response (Error)
+
+```json
+{
+  "success": false,
+  "error": "Workflow not found",
+  "content": "",
+  "toolCalls": []
+}
+```
+
+#### Error Codes
+
+| Status | Error | Description |
+|--------|-------|-------------|
+| 400 | Invalid request | Missing required fields |
+| 401 | Unauthorized | Invalid or missing API key |
+| 404 | Workflow not found | Workflow ID doesn't exist |
+| 500 | Internal error | Server-side failure |
+| 504 | Timeout | Request exceeded timeout |
+
+### Interactive API (Existing - Modified)
+
+The existing `/api/copilot/chat` endpoint continues to work but now uses the orchestrator internally. SSE events forwarded to client remain the same format.
+
+---
+
+## Migration Strategy
+
+### Rollout Plan
+
+```
+Week 1: Phase 1 (Orchestrator)
+├── Day 1-2: Types + SSE Parser
+├── Day 3: Tool Executor
+└── Day 4-5: Handlers + Main Orchestrator
+
+Week 2: Phase 2-3 (Routes)
+├── Day 1: Headless API route
+├── Day 2-3: Wire interactive route
+└── Day 4-5: Testing both modes
+
+Week 3: Phase 4-5 (Cleanup)
+├── Day 1-3: Simplify store
+├── Day 4: Testing
+└── Day 5: Documentation
+```
+
+### Feature Flags
+
+```typescript
+// lib/copilot/config.ts
+
+export const COPILOT_FLAGS = {
+  // Use new orchestrator for interactive mode
+  USE_SERVER_ORCHESTRATOR: process.env.COPILOT_USE_SERVER_ORCHESTRATOR === 'true',
+  
+  // Enable headless API
+  ENABLE_HEADLESS_API: process.env.COPILOT_ENABLE_HEADLESS_API === 'true',
+}
+```
+
+### Rollback Plan
+
+If issues arise:
+1. Set `COPILOT_USE_SERVER_ORCHESTRATOR=false`
+2. Interactive mode falls back to old client-side flow
+3. Headless API returns 503 Service Unavailable
+
+---
+
+## Testing Strategy
+
+### Unit Tests
+
+```
+lib/copilot/orchestrator/
+├── __tests__/
+│   ├── sse-parser.test.ts
+│   ├── sse-handlers.test.ts
+│   ├── tool-executor.test.ts
+│   ├── persistence.test.ts
+│   └── index.test.ts
+```
+
+#### SSE Parser Tests
+
+```typescript
+describe('parseSSEStream', () => {
+  it('parses content events')
+  it('parses tool_call events')
+  it('handles partial lines')
+  it('handles malformed JSON')
+  it('respects abort signal')
+})
+```
+
+#### Tool Executor Tests
+
+```typescript
+describe('executeToolServerSide', () => {
+  it('executes server tools directly')
+  it('executes integration tools with OAuth')
+  it('resolves env var references')
+  it('handles tool not found')
+  it('handles execution errors')
+})
+```
+
+### Integration Tests
+
+```typescript
+describe('orchestrateCopilotRequest', () => {
+  it('handles simple message without tools')
+  it('handles message with single tool call')
+  it('handles message with multiple tool calls')
+  it('handles subagent tool calls')
+  it('handles stream errors')
+  it('respects timeout')
+  it('forwards events to callback')
+})
+```
+
+### E2E Tests
+
+```typescript
+describe('POST /api/v1/copilot/chat', () => {
+  it('returns 401 without API key')
+  it('returns 400 with invalid request')
+  it('executes simple ask query')
+  it('executes workflow modification')
+  it('handles tool execution')
+})
+```
+
+---
+
+## Risks & Mitigations
+
+### Risk 1: Breaking Interactive Mode
+
+**Risk**: Refactoring could break existing interactive copilot.
+
+**Mitigation**:
+- Feature flag for gradual rollout
+- Keep old code path available
+- Extensive E2E testing
+- Staged deployment (internal → beta → production)
+
+### Risk 2: Tool Execution Differences
+
+**Risk**: Tool behavior differs between client and server execution.
+
+**Mitigation**:
+- Reuse existing tool execution logic (same functions)
+- Compare outputs in parallel testing
+- Log discrepancies for investigation
+
+### Risk 3: Performance Regression
+
+**Risk**: Server-side orchestration could be slower.
+
+**Mitigation**:
+- Actually should be faster (no browser round-trips)
+- Benchmark before/after
+- Profile critical paths
+
+### Risk 4: Memory Usage
+
+**Risk**: Server accumulates state during long-running requests.
+
+**Mitigation**:
+- Set reasonable timeouts
+- Clean up context after request
+- Monitor memory in production
+
+### Risk 5: Connection Issues
+
+**Risk**: Long-running SSE connections could drop.
+
+**Mitigation**:
+- Implement reconnection logic
+- Save checkpoints to resume
+- Handle partial completions gracefully
+
+---
+
+## File Inventory
+
+### New Files
+
+| File | Lines | Description |
+|------|-------|-------------|
+| `lib/copilot/orchestrator/types.ts` | ~200 | Type definitions |
+| `lib/copilot/orchestrator/sse-parser.ts` | ~80 | SSE stream parsing |
+| `lib/copilot/orchestrator/sse-handlers.ts` | ~350 | Event handlers |
+| `lib/copilot/orchestrator/tool-executor.ts` | ~300 | Tool execution |
+| `lib/copilot/orchestrator/persistence.ts` | ~120 | DB/Redis operations |
+| `lib/copilot/orchestrator/index.ts` | ~250 | Main orchestrator |
+| `app/api/v1/copilot/chat/route.ts` | ~100 | Headless API |
+| **Total New** | **~1,400** | |
+
+### Modified Files
+
+| File | Change |
+|------|--------|
+| `app/api/copilot/chat/route.ts` | Use orchestrator (optional) |
+| `stores/panel/copilot/store.ts` | Simplify to ~600 lines |
+
+### Deleted Code (from store.ts)
+
+| Section | Lines Removed |
+|---------|---------------|
+| SSE parsing logic | ~150 |
+| `sseHandlers` object | ~750 |
+| `subAgentSSEHandlers` | ~280 |
+| Tool execution logic | ~400 |
+| Client tool instantiators | ~120 |
+| Content block helpers | ~200 |
+| Streaming context | ~100 |
+| **Total Removed** | **~2,000** |
+
+### Net Change
+
+```
+New code:      +1,400 lines (orchestrator module)
+Removed code:  -2,000 lines (from store)
+Modified code: ~200 lines (route changes)
+───────────────────────────────────────
+Net change:    -400 lines (cleaner, more maintainable)
+```
+
+---
+
+## Appendix: Code Extraction Map
+
+### From `stores/panel/copilot/store.ts`
+
+| Source Lines | Destination | Notes |
+|--------------|-------------|-------|
+| 900-1050 (parseSSEStream) | `sse-parser.ts` | Adapt for server |
+| 1120-1867 (sseHandlers) | `sse-handlers.ts` | Remove Zustand deps |
+| 1940-2217 (subAgentSSEHandlers) | `sse-handlers.ts` | Merge with above |
+| 1365-1583 (tool execution) | `tool-executor.ts` | Direct calls |
+| 330-380 (StreamingContext) | `types.ts` | Clean up |
+| 3328-3648 (handleStreamingResponse) | `index.ts` | Main loop |
+
+### From `app/api/copilot/execute-tool/route.ts`
+
+| Source Lines | Destination | Notes |
+|--------------|-------------|-------|
+| 30-247 (POST handler) | `tool-executor.ts` | Extract core logic |
+
+### From `app/api/copilot/confirm/route.ts`
+
+| Source Lines | Destination | Notes |
+|--------------|-------------|-------|
+| 28-89 (updateToolCallStatus) | `persistence.ts` | Redis operations |
+
+---
+
+## Approval & Sign-off
+
+- [ ] Technical review complete
+- [ ] Security review complete
+- [ ] Performance impact assessed
+- [ ] Rollback plan approved
+- [ ] Testing plan approved
+
+---
+
+*Document created: January 2026*
+*Last updated: January 2026*
+