fix(models): memory fixes, provider code typing, cost calculation cleanup (#2515)

* improvement(memory): should not be block scoped

* cleanup provider code

* update other providers

* cleanup fallback code

* remove flaky test

* fix memory

* move streaming fix to right level

* cleanup streaming server

* make memories workspace scoped

* update docs

* fix dedup logic

* fix streaming parsing issue for multiple onStream calls for same block

* fix(provieders): support parallel agent tool calls, consolidate utils

* address greptile comments

* remove all comments

* fixed openrouter response format handling, groq & cerebras response formats

* removed duplicate type

---------

Co-authored-by: waleed <walif6@gmail.com>

packages/db/migrations/0130_bored_master_chief.sql

@@ -0,0 +1,32 @@
+-- Step 1: Add workspace_id as nullable first
+ALTER TABLE "memory" ADD COLUMN "workspace_id" text;
+
+-- Step 2: Backfill workspace_id from workflow's workspace_id
+UPDATE memory m
+SET workspace_id = w.workspace_id
+FROM workflow w
+WHERE m.workflow_id = w.id
+AND w.workspace_id IS NOT NULL;
+
+-- Step 3: Delete rows where workspace_id couldn't be resolved
+DELETE FROM memory WHERE workspace_id IS NULL;
+
+-- Step 4: Now make workspace_id NOT NULL
+ALTER TABLE "memory" ALTER COLUMN "workspace_id" SET NOT NULL;
+
+-- Step 5: Drop old constraint and indexes
+ALTER TABLE "memory" DROP CONSTRAINT IF EXISTS "memory_workflow_id_workflow_id_fk";
+--> statement-breakpoint
+DROP INDEX IF EXISTS "memory_workflow_idx";
+--> statement-breakpoint
+DROP INDEX IF EXISTS "memory_workflow_key_idx";
+
+-- Step 6: Add new foreign key and indexes
+ALTER TABLE "memory" ADD CONSTRAINT "memory_workspace_id_workspace_id_fk" FOREIGN KEY ("workspace_id") REFERENCES "public"."workspace"("id") ON DELETE cascade ON UPDATE no action;
+--> statement-breakpoint
+CREATE INDEX "memory_workspace_idx" ON "memory" USING btree ("workspace_id");
+--> statement-breakpoint
+CREATE UNIQUE INDEX "memory_workspace_key_idx" ON "memory" USING btree ("workspace_id","key");
+
+-- Step 7: Drop old column
+ALTER TABLE "memory" DROP COLUMN IF EXISTS "workflow_id";

packages/db/migrations/meta/_journal.json

@@ -904,6 +904,13 @@
       "when": 1766275541149,
       "tag": "0129_stormy_nightmare",
       "breakpoints": true
+    },
+    {
+      "idx": 130,
+      "version": "7",
+      "when": 1766433914366,
+      "tag": "0130_bored_master_chief",
+      "breakpoints": true
     }
   ]
 }

packages/db/schema.ts

@@ -962,24 +962,21 @@ export const memory = pgTable(
   'memory',
   {
     id: text('id').primaryKey(),
-    workflowId: text('workflow_id').references(() => workflow.id, { onDelete: 'cascade' }),
-    key: text('key').notNull(), // Conversation ID provided by user with format: conversationId:blockId
-    data: jsonb('data').notNull(), // Stores agent messages as array of {role, content} objects
+    workspaceId: text('workspace_id')
+      .notNull()
+      .references(() => workspace.id, { onDelete: 'cascade' }),
+    key: text('key').notNull(),
+    data: jsonb('data').notNull(),
     createdAt: timestamp('created_at').notNull().defaultNow(),
     updatedAt: timestamp('updated_at').notNull().defaultNow(),
     deletedAt: timestamp('deleted_at'),
   },
   (table) => {
     return {
-      // Add index on key for faster lookups
       keyIdx: index('memory_key_idx').on(table.key),
-
-      // Add index on workflowId for faster filtering
-      workflowIdx: index('memory_workflow_idx').on(table.workflowId),
-
-      // Compound unique index to ensure keys are unique per workflow
-      uniqueKeyPerWorkflowIdx: uniqueIndex('memory_workflow_key_idx').on(
-        table.workflowId,
+      workspaceIdx: index('memory_workspace_idx').on(table.workspaceId),
+      uniqueKeyPerWorkspaceIdx: uniqueIndex('memory_workspace_key_idx').on(
+        table.workspaceId,
         table.key
       ),
     }