extracted core chat changes from hackathon/copilot

This commit is contained in:
Swifty
2026-01-07 09:40:13 +01:00
parent 4db13837b9
commit d3dd13fc55
16 changed files with 2779 additions and 82 deletions

View File

@@ -12,7 +12,11 @@ class ChatConfig(BaseSettings):
# OpenAI API Configuration
model: str = Field(
default="qwen/qwen3-235b-a22b-2507", description="Default model to use"
default="anthropic/claude-opus-4.5", description="Default model to use"
)
title_model: str = Field(
default="openai/gpt-4o-mini",
description="Model to use for generating session titles (should be fast/cheap)",
)
api_key: str | None = Field(default=None, description="OpenAI API key")
base_url: str | None = Field(
@@ -72,8 +76,31 @@ class ChatConfig(BaseSettings):
v = "https://openrouter.ai/api/v1"
return v
# Prompt paths for different contexts
PROMPT_PATHS: dict[str, str] = {
"default": "prompts/chat_system.md",
"onboarding": "prompts/onboarding_system.md",
}
def get_system_prompt_for_type(
self, prompt_type: str = "default", **template_vars
) -> str:
"""Load and render a system prompt by type.
Args:
prompt_type: The type of prompt to load ("default" or "onboarding")
**template_vars: Variables to substitute in the template
Returns:
Rendered system prompt string
"""
prompt_path_str = self.PROMPT_PATHS.get(
prompt_type, self.PROMPT_PATHS["default"]
)
return self._load_prompt_from_path(prompt_path_str, **template_vars)
def get_system_prompt(self, **template_vars) -> str:
"""Load and render the system prompt from file.
"""Load and render the default system prompt from file.
Args:
**template_vars: Variables to substitute in the template
@@ -82,9 +109,21 @@ class ChatConfig(BaseSettings):
Rendered system prompt string
"""
return self._load_prompt_from_path(self.system_prompt_path, **template_vars)
def _load_prompt_from_path(self, prompt_path_str: str, **template_vars) -> str:
"""Load and render a system prompt from a given path.
Args:
prompt_path_str: Path to the prompt file relative to chat module
**template_vars: Variables to substitute in the template
Returns:
Rendered system prompt string
"""
# Get the path relative to this module
module_dir = Path(__file__).parent
prompt_path = module_dir / self.system_prompt_path
prompt_path = module_dir / prompt_path_str
# Check for .j2 extension first (Jinja2 template)
j2_path = Path(str(prompt_path) + ".j2")

View File

@@ -0,0 +1,195 @@
"""Database operations for chat sessions."""
import logging
from datetime import UTC, datetime
from typing import Any
from prisma.models import ChatMessage as PrismaChatMessage
from prisma.models import ChatSession as PrismaChatSession
from prisma.types import ChatSessionUpdateInput
from backend.util.json import SafeJson
logger = logging.getLogger(__name__)
async def get_chat_session(session_id: str) -> PrismaChatSession | None:
"""Get a chat session by ID from the database."""
session = await PrismaChatSession.prisma().find_unique(
where={"id": session_id},
include={"Messages": True},
)
if session and session.Messages:
# Sort messages by sequence in Python since Prisma doesn't support order_by in include
session.Messages.sort(key=lambda m: m.sequence)
return session
async def create_chat_session(
session_id: str,
user_id: str | None,
) -> PrismaChatSession:
"""Create a new chat session in the database."""
data = {
"id": session_id,
"userId": user_id,
"credentials": SafeJson({}),
"successfulAgentRuns": SafeJson({}),
"successfulAgentSchedules": SafeJson({}),
}
return await PrismaChatSession.prisma().create(
data=data,
include={"Messages": True},
)
async def update_chat_session(
session_id: str,
credentials: dict[str, Any] | None = None,
successful_agent_runs: dict[str, Any] | None = None,
successful_agent_schedules: dict[str, Any] | None = None,
total_prompt_tokens: int | None = None,
total_completion_tokens: int | None = None,
title: str | None = None,
) -> PrismaChatSession | None:
"""Update a chat session's metadata."""
data: ChatSessionUpdateInput = {"updatedAt": datetime.now(UTC)}
if credentials is not None:
data["credentials"] = SafeJson(credentials)
if successful_agent_runs is not None:
data["successfulAgentRuns"] = SafeJson(successful_agent_runs)
if successful_agent_schedules is not None:
data["successfulAgentSchedules"] = SafeJson(successful_agent_schedules)
if total_prompt_tokens is not None:
data["totalPromptTokens"] = total_prompt_tokens
if total_completion_tokens is not None:
data["totalCompletionTokens"] = total_completion_tokens
if title is not None:
data["title"] = title
session = await PrismaChatSession.prisma().update(
where={"id": session_id},
data=data,
include={"Messages": True},
)
if session and session.Messages:
session.Messages.sort(key=lambda m: m.sequence)
return session
async def add_chat_message(
session_id: str,
role: str,
sequence: int,
content: str | None = None,
name: str | None = None,
tool_call_id: str | None = None,
refusal: str | None = None,
tool_calls: list[dict[str, Any]] | None = None,
function_call: dict[str, Any] | None = None,
) -> PrismaChatMessage:
"""Add a message to a chat session."""
data: dict[str, Any] = {
"Session": {"connect": {"id": session_id}},
"role": role,
"sequence": sequence,
}
if content is not None:
data["content"] = content
if name is not None:
data["name"] = name
if tool_call_id is not None:
data["toolCallId"] = tool_call_id
if refusal is not None:
data["refusal"] = refusal
if tool_calls is not None:
data["toolCalls"] = SafeJson(tool_calls)
if function_call is not None:
data["functionCall"] = SafeJson(function_call)
# Update session's updatedAt timestamp
await PrismaChatSession.prisma().update(
where={"id": session_id},
data={"updatedAt": datetime.now(UTC)},
)
return await PrismaChatMessage.prisma().create(data=data)
async def add_chat_messages_batch(
session_id: str,
messages: list[dict[str, Any]],
start_sequence: int,
) -> list[PrismaChatMessage]:
"""Add multiple messages to a chat session in a batch."""
if not messages:
return []
created_messages = []
for i, msg in enumerate(messages):
data: dict[str, Any] = {
"Session": {"connect": {"id": session_id}},
"role": msg["role"],
"sequence": start_sequence + i,
}
if msg.get("content") is not None:
data["content"] = msg["content"]
if msg.get("name") is not None:
data["name"] = msg["name"]
if msg.get("tool_call_id") is not None:
data["toolCallId"] = msg["tool_call_id"]
if msg.get("refusal") is not None:
data["refusal"] = msg["refusal"]
if msg.get("tool_calls") is not None:
data["toolCalls"] = SafeJson(msg["tool_calls"])
if msg.get("function_call") is not None:
data["functionCall"] = SafeJson(msg["function_call"])
created = await PrismaChatMessage.prisma().create(data=data)
created_messages.append(created)
# Update session's updatedAt timestamp
await PrismaChatSession.prisma().update(
where={"id": session_id},
data={"updatedAt": datetime.now(UTC)},
)
return created_messages
async def get_user_chat_sessions(
user_id: str,
limit: int = 50,
offset: int = 0,
) -> list[PrismaChatSession]:
"""Get chat sessions for a user, ordered by most recent."""
return await PrismaChatSession.prisma().find_many(
where={"userId": user_id},
order={"updatedAt": "desc"},
take=limit,
skip=offset,
)
async def get_user_session_count(user_id: str) -> int:
"""Get the total number of chat sessions for a user."""
return await PrismaChatSession.prisma().count(where={"userId": user_id})
async def delete_chat_session(session_id: str) -> bool:
"""Delete a chat session and all its messages."""
try:
await PrismaChatSession.prisma().delete(where={"id": session_id})
return True
except Exception as e:
logger.error(f"Failed to delete chat session {session_id}: {e}")
return False
async def get_chat_session_message_count(session_id: str) -> int:
"""Get the number of messages in a chat session."""
count = await PrismaChatMessage.prisma().count(where={"sessionId": session_id})
return count

View File

@@ -16,11 +16,15 @@ from openai.types.chat.chat_completion_message_tool_call_param import (
ChatCompletionMessageToolCallParam,
Function,
)
from prisma.models import ChatMessage as PrismaChatMessage
from prisma.models import ChatSession as PrismaChatSession
from pydantic import BaseModel
from backend.data.redis_client import get_redis_async
from backend.util import json
from backend.util.exceptions import RedisError
from . import db as chat_db
from .config import ChatConfig
logger = logging.getLogger(__name__)
@@ -46,6 +50,7 @@ class Usage(BaseModel):
class ChatSession(BaseModel):
session_id: str
user_id: str | None
title: str | None = None
messages: list[ChatMessage]
usage: list[Usage]
credentials: dict[str, dict] = {} # Map of provider -> credential metadata
@@ -59,6 +64,7 @@ class ChatSession(BaseModel):
return ChatSession(
session_id=str(uuid.uuid4()),
user_id=user_id,
title=None,
messages=[],
usage=[],
credentials={},
@@ -66,6 +72,85 @@ class ChatSession(BaseModel):
updated_at=datetime.now(UTC),
)
@staticmethod
def from_prisma(
prisma_session: PrismaChatSession,
prisma_messages: list[PrismaChatMessage] | None = None,
) -> "ChatSession":
"""Convert Prisma models to Pydantic ChatSession."""
messages = []
if prisma_messages:
for msg in prisma_messages:
tool_calls = None
if msg.toolCalls:
tool_calls = (
json.loads(msg.toolCalls)
if isinstance(msg.toolCalls, str)
else msg.toolCalls
)
function_call = None
if msg.functionCall:
function_call = (
json.loads(msg.functionCall)
if isinstance(msg.functionCall, str)
else msg.functionCall
)
messages.append(
ChatMessage(
role=msg.role,
content=msg.content,
name=msg.name,
tool_call_id=msg.toolCallId,
refusal=msg.refusal,
tool_calls=tool_calls,
function_call=function_call,
)
)
# Parse JSON fields from Prisma
credentials = (
json.loads(prisma_session.credentials)
if isinstance(prisma_session.credentials, str)
else prisma_session.credentials or {}
)
successful_agent_runs = (
json.loads(prisma_session.successfulAgentRuns)
if isinstance(prisma_session.successfulAgentRuns, str)
else prisma_session.successfulAgentRuns or {}
)
successful_agent_schedules = (
json.loads(prisma_session.successfulAgentSchedules)
if isinstance(prisma_session.successfulAgentSchedules, str)
else prisma_session.successfulAgentSchedules or {}
)
# Calculate usage from token counts
usage = []
if prisma_session.totalPromptTokens or prisma_session.totalCompletionTokens:
usage.append(
Usage(
prompt_tokens=prisma_session.totalPromptTokens or 0,
completion_tokens=prisma_session.totalCompletionTokens or 0,
total_tokens=(prisma_session.totalPromptTokens or 0)
+ (prisma_session.totalCompletionTokens or 0),
)
)
return ChatSession(
session_id=prisma_session.id,
user_id=prisma_session.userId,
title=prisma_session.title,
messages=messages,
usage=usage,
credentials=credentials,
started_at=prisma_session.createdAt,
updated_at=prisma_session.updatedAt,
successful_agent_runs=successful_agent_runs,
successful_agent_schedules=successful_agent_schedules,
)
def to_openai_messages(self) -> list[ChatCompletionMessageParam]:
messages = []
for message in self.messages:
@@ -155,50 +240,234 @@ class ChatSession(BaseModel):
return messages
async def get_chat_session(
session_id: str,
user_id: str | None,
) -> ChatSession | None:
"""Get a chat session by ID."""
async def _get_session_from_cache(session_id: str) -> ChatSession | None:
"""Get a chat session from Redis cache."""
redis_key = f"chat:session:{session_id}"
async_redis = await get_redis_async()
raw_session: bytes | None = await async_redis.get(redis_key)
if raw_session is None:
logger.warning(f"Session {session_id} not found in Redis")
return None
try:
session = ChatSession.model_validate_json(raw_session)
logger.info(
f"Loading session {session_id} from cache: "
f"message_count={len(session.messages)}, "
f"roles={[m.role for m in session.messages]}"
)
return session
except Exception as e:
logger.error(f"Failed to deserialize session {session_id}: {e}", exc_info=True)
raise RedisError(f"Corrupted session data for {session_id}") from e
async def _cache_session(session: ChatSession) -> None:
"""Cache a chat session in Redis."""
redis_key = f"chat:session:{session.session_id}"
async_redis = await get_redis_async()
await async_redis.setex(redis_key, config.session_ttl, session.model_dump_json())
async def _get_session_from_db(session_id: str) -> ChatSession | None:
"""Get a chat session from the database."""
prisma_session = await chat_db.get_chat_session(session_id)
if not prisma_session:
return None
messages = prisma_session.Messages
logger.info(
f"Loading session {session_id} from DB: "
f"has_messages={messages is not None}, "
f"message_count={len(messages) if messages else 0}, "
f"roles={[m.role for m in messages] if messages else []}"
)
return ChatSession.from_prisma(prisma_session, messages)
async def _save_session_to_db(
session: ChatSession, existing_message_count: int
) -> None:
"""Save or update a chat session in the database."""
# Check if session exists in DB
existing = await chat_db.get_chat_session(session.session_id)
if not existing:
# Create new session
await chat_db.create_chat_session(
session_id=session.session_id,
user_id=session.user_id,
)
existing_message_count = 0
# Calculate total tokens from usage
total_prompt = sum(u.prompt_tokens for u in session.usage)
total_completion = sum(u.completion_tokens for u in session.usage)
# Update session metadata
await chat_db.update_chat_session(
session_id=session.session_id,
credentials=session.credentials,
successful_agent_runs=session.successful_agent_runs,
successful_agent_schedules=session.successful_agent_schedules,
total_prompt_tokens=total_prompt,
total_completion_tokens=total_completion,
)
# Add new messages (only those after existing count)
new_messages = session.messages[existing_message_count:]
if new_messages:
messages_data = []
for msg in new_messages:
messages_data.append(
{
"role": msg.role,
"content": msg.content,
"name": msg.name,
"tool_call_id": msg.tool_call_id,
"refusal": msg.refusal,
"tool_calls": msg.tool_calls,
"function_call": msg.function_call,
}
)
logger.info(
f"Saving {len(new_messages)} new messages to DB for session {session.session_id}: "
f"roles={[m['role'] for m in messages_data]}, "
f"start_sequence={existing_message_count}"
)
await chat_db.add_chat_messages_batch(
session_id=session.session_id,
messages=messages_data,
start_sequence=existing_message_count,
)
async def get_chat_session(
session_id: str,
user_id: str | None,
) -> ChatSession | None:
"""Get a chat session by ID.
Checks Redis cache first, falls back to database if not found.
Caches database results back to Redis.
"""
# Try cache first
try:
session = await _get_session_from_cache(session_id)
if session:
# Verify user ownership
if session.user_id is not None and session.user_id != user_id:
logger.warning(
f"Session {session_id} user id mismatch: {session.user_id} != {user_id}"
)
return None
return session
except RedisError:
logger.warning(f"Cache error for session {session_id}, trying database")
except Exception as e:
logger.warning(f"Unexpected cache error for session {session_id}: {e}")
# Fall back to database
logger.info(f"Session {session_id} not in cache, checking database")
session = await _get_session_from_db(session_id)
if session is None:
logger.warning(f"Session {session_id} not found in cache or database")
return None
# Verify user ownership
if session.user_id is not None and session.user_id != user_id:
logger.warning(
f"Session {session_id} user id mismatch: {session.user_id} != {user_id}"
)
return None
# Cache the session from DB
try:
await _cache_session(session)
logger.info(f"Cached session {session_id} from database")
except Exception as e:
logger.warning(f"Failed to cache session {session_id}: {e}")
return session
async def upsert_chat_session(
session: ChatSession,
) -> ChatSession:
"""Update a chat session with the given messages."""
redis_key = f"chat:session:{session.session_id}"
async_redis = await get_redis_async()
resp = await async_redis.setex(
redis_key, config.session_ttl, session.model_dump_json()
"""Update a chat session in both cache and database."""
# Get existing message count from DB for incremental saves
existing_message_count = await chat_db.get_chat_session_message_count(
session.session_id
)
if not resp:
# Save to database
try:
await _save_session_to_db(session, existing_message_count)
except Exception as e:
logger.error(f"Failed to save session {session.session_id} to database: {e}")
# Continue to cache even if DB fails
# Save to cache
try:
await _cache_session(session)
except Exception as e:
raise RedisError(
f"Failed to persist chat session {session.session_id} to Redis: {resp}"
)
f"Failed to persist chat session {session.session_id} to Redis: {e}"
) from e
return session
async def create_chat_session(user_id: str | None) -> ChatSession:
"""Create a new chat session and persist it."""
session = ChatSession.new(user_id)
# Create in database first
try:
await chat_db.create_chat_session(
session_id=session.session_id,
user_id=user_id,
)
except Exception as e:
logger.error(f"Failed to create session in database: {e}")
# Continue even if DB fails - cache will still work
# Cache the session
try:
await _cache_session(session)
except Exception as e:
logger.warning(f"Failed to cache new session: {e}")
return session
async def get_user_sessions(
user_id: str,
limit: int = 50,
offset: int = 0,
) -> list[ChatSession]:
"""Get all chat sessions for a user from the database."""
prisma_sessions = await chat_db.get_user_chat_sessions(user_id, limit, offset)
sessions = []
for prisma_session in prisma_sessions:
# Convert without messages for listing (lighter weight)
sessions.append(ChatSession.from_prisma(prisma_session, None))
return sessions
async def delete_chat_session(session_id: str) -> bool:
"""Delete a chat session from both cache and database."""
# Delete from cache
try:
redis_key = f"chat:session:{session_id}"
async_redis = await get_redis_async()
await async_redis.delete(redis_key)
except Exception as e:
logger.warning(f"Failed to delete session {session_id} from cache: {e}")
# Delete from database
return await chat_db.delete_chat_session(session_id)

View File

@@ -68,3 +68,50 @@ async def test_chatsession_redis_storage_user_id_mismatch():
s2 = await get_chat_session(s.session_id, None)
assert s2 is None
@pytest.mark.asyncio(loop_scope="session")
async def test_chatsession_db_storage():
"""Test that messages are correctly saved to and loaded from DB (not cache)."""
from backend.data.redis_client import get_redis_async
# Create session with messages including assistant message
s = ChatSession.new(user_id=None)
s.messages = messages # Contains user, assistant, and tool messages
# Upsert to save to both cache and DB
s = await upsert_chat_session(s)
# Clear the Redis cache to force DB load
redis_key = f"chat:session:{s.session_id}"
async_redis = await get_redis_async()
await async_redis.delete(redis_key)
# Load from DB (cache was cleared)
s2 = await get_chat_session(
session_id=s.session_id,
user_id=s.user_id,
)
assert s2 is not None, "Session not found after loading from DB"
assert len(s2.messages) == len(
s.messages
), f"Message count mismatch: expected {len(s.messages)}, got {len(s2.messages)}"
# Verify all roles are present
roles = [m.role for m in s2.messages]
assert "user" in roles, f"User message missing. Roles found: {roles}"
assert "assistant" in roles, f"Assistant message missing. Roles found: {roles}"
assert "tool" in roles, f"Tool message missing. Roles found: {roles}"
# Verify message content
for orig, loaded in zip(s.messages, s2.messages):
assert orig.role == loaded.role, f"Role mismatch: {orig.role} != {loaded.role}"
assert (
orig.content == loaded.content
), f"Content mismatch for {orig.role}: {orig.content} != {loaded.content}"
if orig.tool_calls:
assert (
loaded.tool_calls is not None
), f"Tool calls missing for {orig.role} message"
assert len(orig.tool_calls) == len(loaded.tool_calls)

View File

@@ -1,12 +1,80 @@
You are Otto, an AI Co-Pilot and Forward Deployed Engineer for AutoGPT, an AI Business Automation tool. Your mission is to help users quickly find and set up AutoGPT agents to solve their business problems.
You are Otto, an AI Co-Pilot and Forward Deployed Engineer for AutoGPT, an AI Business Automation tool. Your mission is to help users quickly find, create, and set up AutoGPT agents to solve their business problems.
Here are the functions available to you:
<functions>
1. **find_agent** - Search for agents that solve the user's problem
2. **run_agent** - Run or schedule an agent (automatically handles setup)
**Understanding & Discovery:**
1. **add_understanding** - Save information about the user's business context (use this as you learn about them)
2. **find_agent** - Search the marketplace for pre-built agents that solve the user's problem
3. **find_library_agent** - Search the user's personal library of saved agents
4. **find_block** - Search for individual blocks (building components for agents)
5. **search_platform_docs** - Search AutoGPT documentation for help
**Agent Creation & Editing:**
6. **create_agent** - Create a new custom agent from scratch based on user requirements
7. **edit_agent** - Modify an existing agent (add/remove blocks, change configuration)
**Execution & Output:**
8. **run_agent** - Run or schedule an agent (automatically handles setup)
9. **run_block** - Run a single block directly without creating an agent
10. **agent_output** - Get the output/results from a running or completed agent execution
</functions>
## ALWAYS GET THE USER'S NAME
**This is critical:** If you don't know the user's name, ask for it in your first response. Use a friendly, natural approach:
- "Hi! I'm Otto. What's your name?"
- "Hey there! Before we dive in, what should I call you?"
Once you have their name, immediately save it with `add_understanding(user_name="...")` and use it throughout the conversation.
## BUILDING USER UNDERSTANDING
**If no User Business Context is provided below**, gather information naturally during conversation - don't interrogate them.
**Key information to gather (in priority order):**
1. Their name (ALWAYS first if unknown)
2. Their job title and role
3. Their business/company and industry
4. Pain points and what they want to automate
5. Tools they currently use
**How to gather this information:**
- Ask naturally as part of helping them (e.g., "What's your role?" or "What industry are you in?")
- When they share information, immediately save it using `add_understanding`
- Don't ask all questions at once - spread them across the conversation
- Prioritize understanding their immediate problem first
**Example:**
```
User: "I need help automating my social media"
Otto: I can help with that! I'm Otto - what's your name?
User: "I'm Sarah"
Otto: [calls add_understanding with user_name="Sarah"]
Nice to meet you, Sarah! What's your role - are you a social media manager or business owner?
User: "I'm the marketing director at a fintech startup"
Otto: [calls add_understanding with job_title="Marketing Director", industry="fintech", business_size="startup"]
Great! Let me find social media automation agents for you.
[calls find_agent with query="social media automation marketing"]
```
## WHEN TO USE WHICH TOOL
**Finding existing agents:**
- `find_agent` - Search the marketplace for pre-built agents others have created
- `find_library_agent` - Search agents the user has already saved to their library
**Creating/editing agents:**
- `create_agent` - When user wants a custom agent that doesn't exist, or has specific requirements
- `edit_agent` - When user wants to modify an existing agent (change inputs, add blocks, etc.)
**Running agents:**
- `run_agent` - To execute an agent (handles credentials and inputs automatically)
- `agent_output` - To check the results of a running or completed agent execution
**Direct execution:**
- `run_block` - Run a single block directly without needing a full agent
## HOW run_agent WORKS
The `run_agent` tool automatically handles the entire setup flow:
@@ -21,49 +89,61 @@ Parameters:
- `use_defaults`: Set to `true` to run with default values (only after user confirms)
- `schedule_name` + `cron`: For scheduled execution
## HOW create_agent WORKS
Use `create_agent` when the user wants to build a custom automation:
- Describe what the agent should do
- The tool will create the agent structure with appropriate blocks
- Returns the agent ID for further editing or running
## HOW agent_output WORKS
Use `agent_output` to get results from agent executions:
- Pass the execution_id from a run_agent response
- Returns the current status and any outputs produced
- Useful for checking if an agent has completed and what it produced
## WORKFLOW
1. **find_agent** - Search for agents that solve the user's problem
2. **run_agent** (first call, no inputs) - Get available inputs for the agent
3. **Ask user** what values they want to use OR if they want to use defaults
4. **run_agent** (second call) - Either with `inputs={...}` or `use_defaults=true`
1. **Get their name** - If unknown, ask for it first
2. **Understand context** - Ask 1-2 questions about their problem while helping
3. **Find or create** - Use find_agent for existing solutions, create_agent for custom needs
4. **Set up and run** - Use run_agent to execute, agent_output to get results
## YOUR APPROACH
**Step 1: Understand the Problem**
**Step 1: Greet and Identify**
- If you don't know their name, ask for it
- Be friendly and conversational
**Step 2: Understand the Problem**
- Ask maximum 1-2 targeted questions
- Focus on: What business problem are they solving?
- Move quickly to searching for solutions
- If they want to create/edit an agent, understand what it should do
**Step 2: Find Agents**
- Use `find_agent` immediately with relevant keywords
- Suggest the best option from search results
- Explain briefly how it solves their problem
**Step 3: Find or Create**
- For existing solutions: Use `find_agent` with relevant keywords
- For custom needs: Use `create_agent` with their requirements
- For modifications: Use `edit_agent` on an existing agent
**Step 3: Get Agent Inputs**
- Call `run_agent(username_agent_slug="creator/agent-name")` without inputs
- This returns the available inputs (required and optional)
- Present these to the user and ask what values they want
**Step 4: Execute**
- Call `run_agent` without inputs first to see what's available
- Ask user what values they want or if defaults are okay
- Call `run_agent` again with inputs or `use_defaults=true`
- Use `agent_output` to check results when needed
**Step 4: Run with User's Choice**
- If user provides values: `run_agent(username_agent_slug="...", inputs={...})`
- If user says "use defaults": `run_agent(username_agent_slug="...", use_defaults=true)`
- On success, share the agent link with the user
## USING add_understanding
**For Scheduled Execution:**
- Add `schedule_name` and `cron` parameters
- Example: `run_agent(username_agent_slug="...", inputs={...}, schedule_name="Daily Report", cron="0 9 * * *")`
Call `add_understanding` whenever you learn something about the user:
## FUNCTION CALL FORMAT
**User info:** `user_name`, `job_title`
**Business:** `business_name`, `industry`, `business_size` (1-10, 11-50, 51-200, 201-1000, 1000+), `user_role` (decision maker, implementer, end user)
**Processes:** `key_workflows` (array), `daily_activities` (array)
**Pain points:** `pain_points` (array), `bottlenecks` (array), `manual_tasks` (array), `automation_goals` (array)
**Tools:** `current_software` (array), `existing_automation` (array)
**Other:** `additional_notes`
To call a function, use this exact format:
`<function_call>function_name(parameter="value")</function_call>`
Examples:
- `<function_call>find_agent(query="social media automation")</function_call>`
- `<function_call>run_agent(username_agent_slug="creator/agent-name")</function_call>` (get inputs)
- `<function_call>run_agent(username_agent_slug="creator/agent-name", inputs={"topic": "AI news"})</function_call>`
- `<function_call>run_agent(username_agent_slug="creator/agent-name", use_defaults=true)</function_call>`
Example: `add_understanding(user_name="Sarah", job_title="Marketing Director", industry="fintech")`
## KEY RULES
@@ -73,8 +153,12 @@ Examples:
- Don't run agents without first showing available inputs to the user
- Don't use `use_defaults=true` without user explicitly confirming
- Don't write responses longer than 3 sentences
- Don't interrogate users with many questions - gather info naturally
**What You DO:**
- ALWAYS ask for user's name if you don't have it
- Save user information with `add_understanding` as you learn it
- Use their name when addressing them
- Always call run_agent first without inputs to see what's available
- Ask user what values they want OR if they want to use defaults
- Keep all responses to maximum 3 sentences
@@ -87,18 +171,22 @@ Examples:
## RESPONSE STRUCTURE
Before responding, wrap your analysis in <thinking> tags to systematically plan your approach:
- Check if you know the user's name - if not, ask for it
- Check if you have user context - if not, plan to gather some naturally
- Extract the key business problem or request from the user's message
- Determine what function call (if any) you need to make next
- Plan your response to stay under the 3-sentence maximum
Example interaction:
```
User: "Run the AI news agent for me"
Otto: <function_call>run_agent(username_agent_slug="autogpt/ai-news")</function_call>
[Tool returns: Agent accepts inputs - Required: topic. Optional: num_articles (default: 5)]
Otto: The AI News agent needs a topic. What topic would you like news about, or should I use the defaults?
User: "Use defaults"
Otto: <function_call>run_agent(username_agent_slug="autogpt/ai-news", use_defaults=true)</function_call>
User: "Hi, I want to build an agent that monitors my competitors"
Otto: <thinking>I don't know this user's name. I should ask for it while acknowledging their request.</thinking>
Hi! I'm Otto and I'd love to help you build a competitor monitoring agent. What's your name?
User: "I'm Mike"
Otto: [calls add_understanding with user_name="Mike"]
<thinking>Now I know Mike wants competitor monitoring. I should search for existing agents first.</thinking>
Great to meet you, Mike! Let me search for competitor monitoring agents.
[calls find_agent with query="competitor monitoring analysis"]
```
KEEP ANSWERS TO 3 SENTENCES

View File

@@ -0,0 +1,155 @@
You are Otto, an AI Co-Pilot helping new users get started with AutoGPT, an AI Business Automation platform. Your mission is to welcome them, learn about their needs, and help them run their first successful agent.
Here are the functions available to you:
<functions>
**Understanding & Discovery:**
1. **add_understanding** - Save information about the user's business context (use this as you learn about them)
2. **find_agent** - Search the marketplace for pre-built agents that solve the user's problem
3. **find_library_agent** - Search the user's personal library of saved agents
4. **find_block** - Search for individual blocks (building components for agents)
5. **search_platform_docs** - Search AutoGPT documentation for help
**Agent Creation & Editing:**
6. **create_agent** - Create a new custom agent from scratch based on user requirements
7. **edit_agent** - Modify an existing agent (add/remove blocks, change configuration)
**Execution & Output:**
8. **run_agent** - Run or schedule an agent (automatically handles setup)
9. **run_block** - Run a single block directly without creating an agent
10. **agent_output** - Get the output/results from a running or completed agent execution
</functions>
## YOUR ONBOARDING MISSION
You are guiding a new user through their first experience with AutoGPT. Your goal is to:
1. Welcome them warmly and get their name
2. Learn about them and their business
3. Find or create an agent that solves a real problem for them
4. Get that agent running successfully
5. Celebrate their success and point them to next steps
## PHASE 1: WELCOME & INTRODUCTION
**Start every conversation by:**
- Giving a warm, friendly greeting
- Introducing yourself as Otto, their AI assistant
- Asking for their name immediately
**Example opening:**
```
Hi! I'm Otto, your AI assistant. Welcome to AutoGPT! I'm here to help you set up your first automation. What's your name?
```
Once you have their name, save it immediately with `add_understanding(user_name="...")` and use it throughout.
## PHASE 2: DISCOVERY
**After getting their name, learn about them:**
- What's their role/job title?
- What industry/business are they in?
- What's one thing they'd love to automate?
**Keep it conversational - don't interrogate. Example:**
```
Nice to meet you, Sarah! What do you do for work, and what's one task you wish you could automate?
```
Save everything you learn with `add_understanding`.
## PHASE 3: FIND OR CREATE AN AGENT
**Once you understand their need:**
- Search for existing agents with `find_agent`
- Present the best match and explain how it helps them
- If nothing fits, offer to create a custom agent with `create_agent`
**Be enthusiastic about the solution:**
```
I found a great agent for you! The "Social Media Scheduler" can automatically post to your accounts on a schedule. Want to try it?
```
## PHASE 4: SETUP & RUN
**Guide them through running the agent:**
1. Call `run_agent` without inputs first to see what's needed
2. Explain each input in simple terms
3. Ask what values they want to use
4. Run the agent with their inputs or defaults
**Don't mention credentials** - the UI handles that automatically.
## PHASE 5: CELEBRATE & HANDOFF
**After successful execution:**
- Congratulate them on their first automation!
- Tell them where to find this agent (their Library)
- Mention they can explore more agents in the Marketplace
- Offer to help with anything else
**Example:**
```
You did it! Your first agent is running. You can find it anytime in your Library. Ready to explore more automations?
```
## KEY RULES
**What You DON'T Do:**
- Don't help with login (frontend handles this)
- Don't mention credentials (UI handles automatically)
- Don't run agents without showing inputs first
- Don't use `use_defaults=true` without explicit confirmation
- Don't write responses longer than 3 sentences
- Don't overwhelm with too many questions at once
**What You DO:**
- ALWAYS get the user's name first
- Be warm, encouraging, and celebratory
- Save info with `add_understanding` as you learn it
- Use their name when addressing them
- Keep responses to maximum 3 sentences
- Make them feel successful at each step
## USING add_understanding
Save information as you learn it:
**User info:** `user_name`, `job_title`
**Business:** `business_name`, `industry`, `business_size`, `user_role`
**Pain points:** `pain_points`, `manual_tasks`, `automation_goals`
**Tools:** `current_software`
Example: `add_understanding(user_name="Sarah", job_title="Marketing Manager", automation_goals=["social media scheduling"])`
## HOW run_agent WORKS
1. **First call** (no inputs) → Shows available inputs
2. **Credentials** → UI handles automatically (don't mention)
3. **Execution** → Run with `inputs={...}` or `use_defaults=true`
## RESPONSE STRUCTURE
Before responding, plan your approach in <thinking> tags:
- What phase am I in? (Welcome/Discovery/Find/Setup/Celebrate)
- Do I know their name? If not, ask for it
- What's the next step to move them forward?
- Keep response under 3 sentences
**Example flow:**
```
User: "Hi"
Otto: <thinking>Phase 1 - I need to welcome them and get their name.</thinking>
Hi! I'm Otto, welcome to AutoGPT! I'm here to help you set up your first automation - what's your name?
User: "I'm Alex"
Otto: [calls add_understanding with user_name="Alex"]
<thinking>Got their name. Phase 2 - learn about them.</thinking>
Great to meet you, Alex! What do you do for work, and what's one task you'd love to automate?
User: "I run an e-commerce store and spend hours on customer support emails"
Otto: [calls add_understanding with industry="e-commerce", pain_points=["customer support emails"]]
<thinking>Phase 3 - search for agents.</thinking>
[calls find_agent with query="customer support email automation"]
```
KEEP ANSWERS TO 3 SENTENCES - Be warm, helpful, and focused on their success!

View File

@@ -26,6 +26,14 @@ router = APIRouter(
# ========== Request/Response Models ==========
class StreamChatRequest(BaseModel):
"""Request model for streaming chat with optional context."""
message: str
is_user_message: bool = True
context: dict[str, str] | None = None # {url: str, content: str}
class CreateSessionResponse(BaseModel):
"""Response model containing information on a newly created chat session."""
@@ -44,9 +52,64 @@ class SessionDetailResponse(BaseModel):
messages: list[dict]
class SessionSummaryResponse(BaseModel):
"""Response model for a session summary (without messages)."""
id: str
created_at: str
updated_at: str
title: str | None = None
class ListSessionsResponse(BaseModel):
"""Response model for listing chat sessions."""
sessions: list[SessionSummaryResponse]
total: int
# ========== Routes ==========
@router.get(
"/sessions",
dependencies=[Security(auth.requires_user)],
)
async def list_sessions(
user_id: Annotated[str, Security(auth.get_user_id)],
limit: int = Query(default=50, ge=1, le=100),
offset: int = Query(default=0, ge=0),
) -> ListSessionsResponse:
"""
List chat sessions for the authenticated user.
Returns a paginated list of chat sessions belonging to the current user,
ordered by most recently updated.
Args:
user_id: The authenticated user's ID.
limit: Maximum number of sessions to return (1-100).
offset: Number of sessions to skip for pagination.
Returns:
ListSessionsResponse: List of session summaries and total count.
"""
sessions = await chat_service.get_user_sessions(user_id, limit, offset)
return ListSessionsResponse(
sessions=[
SessionSummaryResponse(
id=session.session_id,
created_at=session.started_at.isoformat(),
updated_at=session.updated_at.isoformat(),
title=None, # TODO: Add title support
)
for session in sessions
],
total=len(sessions),
)
@router.post(
"/sessions",
)
@@ -102,26 +165,89 @@ async def get_session(
session = await chat_service.get_session(session_id, user_id)
if not session:
raise NotFoundError(f"Session {session_id} not found")
messages = [message.model_dump() for message in session.messages]
logger.info(
f"Returning session {session_id}: "
f"message_count={len(messages)}, "
f"roles={[m.get('role') for m in messages]}"
)
return SessionDetailResponse(
id=session.session_id,
created_at=session.started_at.isoformat(),
updated_at=session.updated_at.isoformat(),
user_id=session.user_id or None,
messages=[message.model_dump() for message in session.messages],
messages=messages,
)
@router.post(
"/sessions/{session_id}/stream",
)
async def stream_chat_post(
session_id: str,
request: StreamChatRequest,
user_id: str | None = Depends(auth.get_user_id),
):
"""
Stream chat responses for a session (POST with context support).
Streams the AI/completion responses in real time over Server-Sent Events (SSE), including:
- Text fragments as they are generated
- Tool call UI elements (if invoked)
- Tool execution results
Args:
session_id: The chat session identifier to associate with the streamed messages.
request: Request body containing message, is_user_message, and optional context.
user_id: Optional authenticated user ID.
Returns:
StreamingResponse: SSE-formatted response chunks.
"""
# Validate session exists before starting the stream
# This prevents errors after the response has already started
session = await chat_service.get_session(session_id, user_id)
if not session:
raise NotFoundError(f"Session {session_id} not found. ")
if session.user_id is None and user_id is not None:
session = await chat_service.assign_user_to_session(session_id, user_id)
async def event_generator() -> AsyncGenerator[str, None]:
async for chunk in chat_service.stream_chat_completion(
session_id,
request.message,
is_user_message=request.is_user_message,
user_id=user_id,
session=session, # Pass pre-fetched session to avoid double-fetch
context=request.context,
):
yield chunk.to_sse()
return StreamingResponse(
event_generator(),
media_type="text/event-stream",
headers={
"Cache-Control": "no-cache",
"Connection": "keep-alive",
"X-Accel-Buffering": "no", # Disable nginx buffering
},
)
@router.get(
"/sessions/{session_id}/stream",
)
async def stream_chat(
async def stream_chat_get(
session_id: str,
message: Annotated[str, Query(min_length=1, max_length=10000)],
user_id: str | None = Depends(auth.get_user_id),
is_user_message: bool = Query(default=True),
):
"""
Stream chat responses for a session.
Stream chat responses for a session (GET - legacy endpoint).
Streams the AI/completion responses in real time over Server-Sent Events (SSE), including:
- Text fragments as they are generated
@@ -193,6 +319,133 @@ async def session_assign_user(
return {"status": "ok"}
# ========== Onboarding Routes ==========
# These routes use a specialized onboarding system prompt
@router.post(
"/onboarding/sessions",
)
async def create_onboarding_session(
user_id: Annotated[str | None, Depends(auth.get_user_id)],
) -> CreateSessionResponse:
"""
Create a new onboarding chat session.
Initiates a new chat session specifically for user onboarding,
using a specialized prompt that guides users through their first
experience with AutoGPT.
Args:
user_id: The optional authenticated user ID parsed from the JWT.
Returns:
CreateSessionResponse: Details of the created onboarding session.
"""
logger.info(
f"Creating onboarding session with user_id: "
f"...{user_id[-8:] if user_id and len(user_id) > 8 else '<redacted>'}"
)
session = await chat_service.create_chat_session(user_id)
return CreateSessionResponse(
id=session.session_id,
created_at=session.started_at.isoformat(),
user_id=session.user_id or None,
)
@router.get(
"/onboarding/sessions/{session_id}",
)
async def get_onboarding_session(
session_id: str,
user_id: Annotated[str | None, Depends(auth.get_user_id)],
) -> SessionDetailResponse:
"""
Retrieve the details of an onboarding chat session.
Args:
session_id: The unique identifier for the onboarding session.
user_id: The optional authenticated user ID.
Returns:
SessionDetailResponse: Details for the requested session.
"""
session = await chat_service.get_session(session_id, user_id)
if not session:
raise NotFoundError(f"Session {session_id} not found")
messages = [message.model_dump() for message in session.messages]
logger.info(
f"Returning onboarding session {session_id}: "
f"message_count={len(messages)}, "
f"roles={[m.get('role') for m in messages]}"
)
return SessionDetailResponse(
id=session.session_id,
created_at=session.started_at.isoformat(),
updated_at=session.updated_at.isoformat(),
user_id=session.user_id or None,
messages=messages,
)
@router.post(
"/onboarding/sessions/{session_id}/stream",
)
async def stream_onboarding_chat(
session_id: str,
request: StreamChatRequest,
user_id: str | None = Depends(auth.get_user_id),
):
"""
Stream onboarding chat responses for a session.
Uses the specialized onboarding system prompt to guide new users
through their first experience with AutoGPT. Streams AI responses
in real time over Server-Sent Events (SSE).
Args:
session_id: The onboarding session identifier.
request: Request body containing message and optional context.
user_id: Optional authenticated user ID.
Returns:
StreamingResponse: SSE-formatted response chunks.
"""
session = await chat_service.get_session(session_id, user_id)
if not session:
raise NotFoundError(f"Session {session_id} not found.")
if session.user_id is None and user_id is not None:
session = await chat_service.assign_user_to_session(session_id, user_id)
async def event_generator() -> AsyncGenerator[str, None]:
async for chunk in chat_service.stream_chat_completion(
session_id,
request.message,
is_user_message=request.is_user_message,
user_id=user_id,
session=session,
context=request.context,
prompt_type="onboarding", # Use onboarding system prompt
):
yield chunk.to_sse()
return StreamingResponse(
event_generator(),
media_type="text/event-stream",
headers={
"Cache-Control": "no-cache",
"Connection": "keep-alive",
"X-Accel-Buffering": "no",
},
)
# ========== Health Check ==========

View File

@@ -7,8 +7,13 @@ import orjson
from openai import AsyncOpenAI
from openai.types.chat import ChatCompletionChunk, ChatCompletionToolParam
from backend.data.understanding import (
format_understanding_for_prompt,
get_business_understanding,
)
from backend.util.exceptions import NotFoundError
from . import db as chat_db
from .config import ChatConfig
from .model import (
ChatMessage,
@@ -17,6 +22,9 @@ from .model import (
get_chat_session,
upsert_chat_session,
)
from .model import (
create_chat_session as model_create_chat_session,
)
from .response_model import (
StreamBaseResponse,
StreamEnd,
@@ -36,15 +44,109 @@ config = ChatConfig()
client = AsyncOpenAI(api_key=config.api_key, base_url=config.base_url)
async def _is_first_session(user_id: str) -> bool:
"""Check if this is the user's first chat session.
Returns True if the user has 1 or fewer sessions (meaning this is their first).
"""
try:
session_count = await chat_db.get_user_session_count(user_id)
return session_count <= 1
except Exception as e:
logger.warning(f"Failed to check session count for user {user_id}: {e}")
return False # Default to non-onboarding if we can't check
async def _build_system_prompt(
user_id: str | None, prompt_type: str = "default"
) -> str:
"""Build the full system prompt including business understanding if available.
Args:
user_id: The user ID for fetching business understanding
prompt_type: The type of prompt to load ("default" or "onboarding")
If "default" and this is the user's first session, will use "onboarding" instead.
Returns:
The full system prompt with business understanding context if available
"""
# Auto-detect: if using default prompt and this is user's first session, use onboarding
effective_prompt_type = prompt_type
if prompt_type == "default" and user_id:
if await _is_first_session(user_id):
logger.info("First session detected for user, using onboarding prompt")
effective_prompt_type = "onboarding"
# Start with the base system prompt for the specified type
base_prompt = config.get_system_prompt_for_type(effective_prompt_type)
# If user is authenticated, try to fetch their business understanding
if user_id:
try:
understanding = await get_business_understanding(user_id)
if understanding:
context = format_understanding_for_prompt(understanding)
if context:
return (
f"{base_prompt}\n\n---\n\n"
f"{context}\n\n"
"Use this context to provide more personalized recommendations "
"and to better understand the user's business needs when "
"suggesting agents and automations."
)
except Exception as e:
logger.warning(f"Failed to fetch business understanding: {e}")
return base_prompt
async def _generate_session_title(message: str) -> str | None:
"""Generate a concise title for a chat session based on the first message.
Args:
message: The first user message in the session
Returns:
A short title (3-6 words) or None if generation fails
"""
try:
response = await client.chat.completions.create(
model=config.title_model,
messages=[
{
"role": "system",
"content": (
"Generate a very short title (3-6 words) for a chat conversation "
"based on the user's first message. The title should capture the "
"main topic or intent. Return ONLY the title, no quotes or punctuation."
),
},
{"role": "user", "content": message[:500]}, # Limit input length
],
max_tokens=20,
temperature=0.7,
)
title = response.choices[0].message.content
if title:
# Clean up the title
title = title.strip().strip("\"'")
# Limit length
if len(title) > 50:
title = title[:47] + "..."
return title
return None
except Exception as e:
logger.warning(f"Failed to generate session title: {e}")
return None
async def create_chat_session(
user_id: str | None = None,
) -> ChatSession:
"""
Create a new chat session and persist it to the database.
"""
session = ChatSession.new(user_id)
# Persist the session immediately so it can be used for streaming
return await upsert_chat_session(session)
return await model_create_chat_session(user_id)
async def get_session(
@@ -57,6 +159,19 @@ async def get_session(
return await get_chat_session(session_id, user_id)
async def get_user_sessions(
user_id: str,
limit: int = 50,
offset: int = 0,
) -> list[ChatSession]:
"""
Get all chat sessions for a user.
"""
from .model import get_user_sessions as model_get_user_sessions
return await model_get_user_sessions(user_id, limit, offset)
async def assign_user_to_session(
session_id: str,
user_id: str,
@@ -78,6 +193,8 @@ async def stream_chat_completion(
user_id: str | None = None,
retry_count: int = 0,
session: ChatSession | None = None,
context: dict[str, str] | None = None, # {url: str, content: str}
prompt_type: str = "default",
) -> AsyncGenerator[StreamBaseResponse, None]:
"""Main entry point for streaming chat completions with database handling.
@@ -89,6 +206,7 @@ async def stream_chat_completion(
user_message: User's input message
user_id: User ID for authentication (None for anonymous)
session: Optional pre-loaded session object (for recursive calls to avoid Redis refetch)
prompt_type: The type of prompt to use ("default" or "onboarding")
Yields:
StreamBaseResponse objects formatted as SSE
@@ -121,9 +239,18 @@ async def stream_chat_completion(
)
if message:
# Build message content with context if provided
message_content = message
if context and context.get("url") and context.get("content"):
context_text = f"Page URL: {context['url']}\n\nPage Content:\n{context['content']}\n\n---\n\nUser Message: {message}"
message_content = context_text
logger.info(
f"Including page context: URL={context['url']}, content_length={len(context['content'])}"
)
session.messages.append(
ChatMessage(
role="user" if is_user_message else "assistant", content=message
role="user" if is_user_message else "assistant", content=message_content
)
)
logger.info(
@@ -141,6 +268,32 @@ async def stream_chat_completion(
session = await upsert_chat_session(session)
assert session, "Session not found"
# Generate title for new sessions on first user message (non-blocking)
# Check: is_user_message, no title yet, and this is the first user message
if is_user_message and message and not session.title:
user_messages = [m for m in session.messages if m.role == "user"]
if len(user_messages) == 1:
# First user message - generate title in background
import asyncio
async def _update_title():
try:
title = await _generate_session_title(message)
if title:
session.title = title
await upsert_chat_session(session)
logger.info(
f"Generated title for session {session_id}: {title}"
)
except Exception as e:
logger.warning(f"Failed to update session title: {e}")
# Fire and forget - don't block the chat response
asyncio.create_task(_update_title())
# Build system prompt with business understanding
system_prompt = await _build_system_prompt(user_id, prompt_type)
assistant_response = ChatMessage(
role="assistant",
content="",
@@ -159,6 +312,7 @@ async def stream_chat_completion(
async for chunk in _stream_chat_chunks(
session=session,
tools=tools,
system_prompt=system_prompt,
):
if isinstance(chunk, StreamTextChunk):
@@ -279,6 +433,7 @@ async def stream_chat_completion(
user_id=user_id,
retry_count=retry_count + 1,
session=session,
prompt_type=prompt_type,
):
yield chunk
return # Exit after retry to avoid double-saving in finally block
@@ -324,6 +479,7 @@ async def stream_chat_completion(
session_id=session.session_id,
user_id=user_id,
session=session, # Pass session object to avoid Redis refetch
prompt_type=prompt_type,
):
yield chunk
@@ -331,6 +487,7 @@ async def stream_chat_completion(
async def _stream_chat_chunks(
session: ChatSession,
tools: list[ChatCompletionToolParam],
system_prompt: str | None = None,
) -> AsyncGenerator[StreamBaseResponse, None]:
"""
Pure streaming function for OpenAI chat completions with tool calling.
@@ -338,9 +495,9 @@ async def _stream_chat_chunks(
This function is database-agnostic and focuses only on streaming logic.
Args:
messages: Conversation context as ChatCompletionMessageParam list
session_id: Session ID
user_id: User ID for tool execution
session: Chat session with conversation history
tools: Available tools for the model
system_prompt: System prompt to prepend to messages
Yields:
SSE formatted JSON response objects
@@ -350,6 +507,17 @@ async def _stream_chat_chunks(
logger.info("Starting pure chat stream")
# Build messages with system prompt prepended
messages = session.to_openai_messages()
if system_prompt:
from openai.types.chat import ChatCompletionSystemMessageParam
system_message = ChatCompletionSystemMessageParam(
role="system",
content=system_prompt,
)
messages = [system_message] + messages
# Loop to handle tool calls and continue conversation
while True:
try:
@@ -358,7 +526,7 @@ async def _stream_chat_chunks(
# Create the stream with proper types
stream = await client.chat.completions.create(
model=model,
messages=session.to_openai_messages(),
messages=messages,
tools=tools,
tool_choice="auto",
stream=True,
@@ -502,8 +670,12 @@ async def _yield_tool_call(
"""
logger.info(f"Yielding tool call: {tool_calls[yield_idx]}")
# Parse tool call arguments - exceptions will propagate to caller
arguments = orjson.loads(tool_calls[yield_idx]["function"]["arguments"])
# Parse tool call arguments - handle empty arguments gracefully
raw_arguments = tool_calls[yield_idx]["function"]["arguments"]
if raw_arguments:
arguments = orjson.loads(raw_arguments)
else:
arguments = {}
yield StreamToolCall(
tool_id=tool_calls[yield_idx]["id"],

View File

@@ -4,21 +4,45 @@ from openai.types.chat import ChatCompletionToolParam
from backend.api.features.chat.model import ChatSession
from .add_understanding import AddUnderstandingTool
from .agent_output import AgentOutputTool
from .base import BaseTool
from .create_agent import CreateAgentTool
from .edit_agent import EditAgentTool
from .find_agent import FindAgentTool
from .find_block import FindBlockTool
from .find_library_agent import FindLibraryAgentTool
from .run_agent import RunAgentTool
from .run_block import RunBlockTool
from .search_docs import SearchDocsTool
if TYPE_CHECKING:
from backend.api.features.chat.response_model import StreamToolExecutionResult
# Initialize tool instances
add_understanding_tool = AddUnderstandingTool()
create_agent_tool = CreateAgentTool()
edit_agent_tool = EditAgentTool()
find_agent_tool = FindAgentTool()
find_block_tool = FindBlockTool()
find_library_agent_tool = FindLibraryAgentTool()
run_agent_tool = RunAgentTool()
run_block_tool = RunBlockTool()
search_docs_tool = SearchDocsTool()
agent_output_tool = AgentOutputTool()
# Export tools as OpenAI format
tools: list[ChatCompletionToolParam] = [
add_understanding_tool.as_openai_tool(),
create_agent_tool.as_openai_tool(),
edit_agent_tool.as_openai_tool(),
find_agent_tool.as_openai_tool(),
find_block_tool.as_openai_tool(),
find_library_agent_tool.as_openai_tool(),
run_agent_tool.as_openai_tool(),
run_block_tool.as_openai_tool(),
search_docs_tool.as_openai_tool(),
agent_output_tool.as_openai_tool(),
]
@@ -31,8 +55,16 @@ async def execute_tool(
) -> "StreamToolExecutionResult":
tool_map: dict[str, BaseTool] = {
"add_understanding": add_understanding_tool,
"create_agent": create_agent_tool,
"edit_agent": edit_agent_tool,
"find_agent": find_agent_tool,
"find_block": find_block_tool,
"find_library_agent": find_library_agent_tool,
"run_agent": run_agent_tool,
"run_block": run_block_tool,
"search_platform_docs": search_docs_tool,
"agent_output": agent_output_tool,
}
if tool_name not in tool_map:
raise ValueError(f"Tool {tool_name} not found")

View File

@@ -0,0 +1,455 @@
"""Tool for retrieving agent execution outputs from user's library."""
import logging
import re
from datetime import datetime, timedelta, timezone
from typing import Any
from pydantic import BaseModel, field_validator
from backend.api.features.chat.model import ChatSession
from backend.api.features.library import db as library_db
from backend.api.features.library.model import LibraryAgent
from backend.data import execution as execution_db
from backend.data.execution import ExecutionStatus, GraphExecution, GraphExecutionMeta
from .base import BaseTool
from .models import (
AgentOutputResponse,
ErrorResponse,
ExecutionOutputInfo,
NoResultsResponse,
ToolResponseBase,
)
from .utils import fetch_graph_from_store_slug
logger = logging.getLogger(__name__)
class AgentOutputInput(BaseModel):
"""Input parameters for the agent_output tool."""
agent_name: str = ""
library_agent_id: str = ""
store_slug: str = ""
execution_id: str = ""
run_time: str = "latest"
@field_validator(
"agent_name",
"library_agent_id",
"store_slug",
"execution_id",
"run_time",
mode="before",
)
@classmethod
def strip_strings(cls, v: Any) -> Any:
"""Strip whitespace from string fields."""
return v.strip() if isinstance(v, str) else v
def parse_time_expression(
time_expr: str | None,
) -> tuple[datetime | None, datetime | None]:
"""
Parse time expression into datetime range (start, end).
Supports:
- "latest" or None -> returns (None, None) to get most recent
- "yesterday" -> 24h window for yesterday
- "today" -> Today from midnight
- "last week" / "last 7 days" -> 7 day window
- "last month" / "last 30 days" -> 30 day window
- ISO date "YYYY-MM-DD" -> 24h window for that date
"""
if not time_expr or time_expr.lower() == "latest":
return None, None
now = datetime.now(timezone.utc)
expr = time_expr.lower().strip()
# Relative expressions
if expr == "yesterday":
end = now.replace(hour=0, minute=0, second=0, microsecond=0)
start = end - timedelta(days=1)
return start, end
if expr in ("last week", "last 7 days"):
return now - timedelta(days=7), now
if expr in ("last month", "last 30 days"):
return now - timedelta(days=30), now
if expr == "today":
start = now.replace(hour=0, minute=0, second=0, microsecond=0)
return start, now
# Try ISO date format (YYYY-MM-DD)
date_match = re.match(r"^(\d{4})-(\d{2})-(\d{2})$", expr)
if date_match:
year, month, day = map(int, date_match.groups())
start = datetime(year, month, day, 0, 0, 0, tzinfo=timezone.utc)
end = start + timedelta(days=1)
return start, end
# Try ISO datetime
try:
parsed = datetime.fromisoformat(expr.replace("Z", "+00:00"))
if parsed.tzinfo is None:
parsed = parsed.replace(tzinfo=timezone.utc)
# Return +/- 1 hour window around the specified time
return parsed - timedelta(hours=1), parsed + timedelta(hours=1)
except ValueError:
pass
# Fallback: treat as "latest"
return None, None
class AgentOutputTool(BaseTool):
"""Tool for retrieving execution outputs from user's library agents."""
@property
def name(self) -> str:
return "agent_output"
@property
def description(self) -> str:
return """Retrieve execution outputs from agents in the user's library.
Identify the agent using one of:
- agent_name: Fuzzy search in user's library
- library_agent_id: Exact library agent ID
- store_slug: Marketplace format 'username/agent-name'
Select which run to retrieve using:
- execution_id: Specific execution ID
- run_time: 'latest' (default), 'yesterday', 'last week', or ISO date 'YYYY-MM-DD'
"""
@property
def parameters(self) -> dict[str, Any]:
return {
"type": "object",
"properties": {
"agent_name": {
"type": "string",
"description": "Agent name to search for in user's library (fuzzy match)",
},
"library_agent_id": {
"type": "string",
"description": "Exact library agent ID",
},
"store_slug": {
"type": "string",
"description": "Marketplace identifier: 'username/agent-slug'",
},
"execution_id": {
"type": "string",
"description": "Specific execution ID to retrieve",
},
"run_time": {
"type": "string",
"description": (
"Time filter: 'latest', 'yesterday', 'last week', or 'YYYY-MM-DD'"
),
},
},
"required": [],
}
@property
def requires_auth(self) -> bool:
return True
async def _resolve_agent(
self,
user_id: str,
agent_name: str | None,
library_agent_id: str | None,
store_slug: str | None,
) -> tuple[LibraryAgent | None, str | None]:
"""
Resolve agent from provided identifiers.
Returns (library_agent, error_message).
"""
# Priority 1: Exact library agent ID
if library_agent_id:
try:
agent = await library_db.get_library_agent(library_agent_id, user_id)
return agent, None
except Exception as e:
logger.warning(f"Failed to get library agent by ID: {e}")
return None, f"Library agent '{library_agent_id}' not found"
# Priority 2: Store slug (username/agent-name)
if store_slug and "/" in store_slug:
username, agent_slug = store_slug.split("/", 1)
graph, _ = await fetch_graph_from_store_slug(username, agent_slug)
if not graph:
return None, f"Agent '{store_slug}' not found in marketplace"
# Find in user's library by graph_id
agent = await library_db.get_library_agent_by_graph_id(user_id, graph.id)
if not agent:
return (
None,
f"Agent '{store_slug}' is not in your library. "
"Add it first to see outputs.",
)
return agent, None
# Priority 3: Fuzzy name search in library
if agent_name:
try:
response = await library_db.list_library_agents(
user_id=user_id,
search_term=agent_name,
page_size=5,
)
if not response.agents:
return (
None,
f"No agents matching '{agent_name}' found in your library",
)
# Return best match (first result from search)
return response.agents[0], None
except Exception as e:
logger.error(f"Error searching library agents: {e}")
return None, f"Error searching for agent: {e}"
return (
None,
"Please specify an agent name, library_agent_id, or store_slug",
)
async def _get_execution(
self,
user_id: str,
graph_id: str,
execution_id: str | None,
time_start: datetime | None,
time_end: datetime | None,
) -> tuple[GraphExecution | None, list[GraphExecutionMeta], str | None]:
"""
Fetch execution(s) based on filters.
Returns (single_execution, available_executions_meta, error_message).
"""
# If specific execution_id provided, fetch it directly
if execution_id:
execution = await execution_db.get_graph_execution(
user_id=user_id,
execution_id=execution_id,
include_node_executions=False,
)
if not execution:
return None, [], f"Execution '{execution_id}' not found"
return execution, [], None
# Get completed executions with time filters
executions = await execution_db.get_graph_executions(
graph_id=graph_id,
user_id=user_id,
statuses=[ExecutionStatus.COMPLETED],
created_time_gte=time_start,
created_time_lte=time_end,
limit=10,
)
if not executions:
return None, [], None # No error, just no executions
# If only one execution, fetch full details
if len(executions) == 1:
full_execution = await execution_db.get_graph_execution(
user_id=user_id,
execution_id=executions[0].id,
include_node_executions=False,
)
return full_execution, [], None
# Multiple executions - return latest with full details, plus list of available
full_execution = await execution_db.get_graph_execution(
user_id=user_id,
execution_id=executions[0].id,
include_node_executions=False,
)
return full_execution, executions, None
def _build_response(
self,
agent: LibraryAgent,
execution: GraphExecution | None,
available_executions: list[GraphExecutionMeta],
session_id: str | None,
) -> AgentOutputResponse:
"""Build the response based on execution data."""
library_agent_link = f"/library/agents/{agent.id}"
if not execution:
return AgentOutputResponse(
message=f"No completed executions found for agent '{agent.name}'",
session_id=session_id,
agent_name=agent.name,
agent_id=agent.graph_id,
library_agent_id=agent.id,
library_agent_link=library_agent_link,
total_executions=0,
)
execution_info = ExecutionOutputInfo(
execution_id=execution.id,
status=execution.status.value,
started_at=execution.started_at,
ended_at=execution.ended_at,
outputs=dict(execution.outputs),
inputs_summary=execution.inputs if execution.inputs else None,
)
available_list = None
if len(available_executions) > 1:
available_list = [
{
"id": e.id,
"status": e.status.value,
"started_at": e.started_at.isoformat() if e.started_at else None,
}
for e in available_executions[:5]
]
message = f"Found execution outputs for agent '{agent.name}'"
if len(available_executions) > 1:
message += (
f". Showing latest of {len(available_executions)} matching executions."
)
return AgentOutputResponse(
message=message,
session_id=session_id,
agent_name=agent.name,
agent_id=agent.graph_id,
library_agent_id=agent.id,
library_agent_link=library_agent_link,
execution=execution_info,
available_executions=available_list,
total_executions=len(available_executions) if available_executions else 1,
)
async def _execute(
self,
user_id: str | None,
session: ChatSession,
**kwargs,
) -> ToolResponseBase:
"""Execute the agent_output tool."""
session_id = session.session_id
# Parse and validate input
try:
input_data = AgentOutputInput(**kwargs)
except Exception as e:
logger.error(f"Invalid input: {e}")
return ErrorResponse(
message="Invalid input parameters",
error=str(e),
session_id=session_id,
)
# Ensure user_id is present (should be guaranteed by requires_auth)
if not user_id:
return ErrorResponse(
message="User authentication required",
session_id=session_id,
)
# Check if at least one identifier is provided
if not any(
[
input_data.agent_name,
input_data.library_agent_id,
input_data.store_slug,
input_data.execution_id,
]
):
return ErrorResponse(
message=(
"Please specify at least one of: agent_name, "
"library_agent_id, store_slug, or execution_id"
),
session_id=session_id,
)
# If only execution_id provided, we need to find the agent differently
if (
input_data.execution_id
and not input_data.agent_name
and not input_data.library_agent_id
and not input_data.store_slug
):
# Fetch execution directly to get graph_id
execution = await execution_db.get_graph_execution(
user_id=user_id,
execution_id=input_data.execution_id,
include_node_executions=False,
)
if not execution:
return ErrorResponse(
message=f"Execution '{input_data.execution_id}' not found",
session_id=session_id,
)
# Find library agent by graph_id
agent = await library_db.get_library_agent_by_graph_id(
user_id, execution.graph_id
)
if not agent:
return NoResultsResponse(
message=(
f"Execution found but agent not in your library. "
f"Graph ID: {execution.graph_id}"
),
session_id=session_id,
suggestions=["Add the agent to your library to see more details"],
)
return self._build_response(agent, execution, [], session_id)
# Resolve agent from identifiers
agent, error = await self._resolve_agent(
user_id=user_id,
agent_name=input_data.agent_name or None,
library_agent_id=input_data.library_agent_id or None,
store_slug=input_data.store_slug or None,
)
if error or not agent:
return NoResultsResponse(
message=error or "Agent not found",
session_id=session_id,
suggestions=[
"Check the agent name or ID",
"Make sure the agent is in your library",
],
)
# Parse time expression
time_start, time_end = parse_time_expression(input_data.run_time)
# Fetch execution(s)
execution, available_executions, exec_error = await self._get_execution(
user_id=user_id,
graph_id=agent.graph_id,
execution_id=input_data.execution_id or None,
time_start=time_start,
time_end=time_end,
)
if exec_error:
return ErrorResponse(
message=exec_error,
session_id=session_id,
)
return self._build_response(agent, execution, available_executions, session_id)

View File

@@ -0,0 +1,279 @@
"""CreateAgentTool - Creates agents from natural language descriptions."""
import logging
from typing import Any
from backend.api.features.chat.model import ChatSession
from .agent_generator import (
apply_all_fixes,
decompose_goal,
generate_agent,
get_blocks_info,
save_agent_to_library,
validate_agent,
)
from .base import BaseTool
from .models import (
AgentPreviewResponse,
AgentSavedResponse,
ClarificationNeededResponse,
ClarifyingQuestion,
ErrorResponse,
ToolResponseBase,
)
logger = logging.getLogger(__name__)
# Maximum retries for agent generation with validation feedback
MAX_GENERATION_RETRIES = 2
class CreateAgentTool(BaseTool):
"""Tool for creating agents from natural language descriptions."""
@property
def name(self) -> str:
return "create_agent"
@property
def description(self) -> str:
return (
"Create a new agent workflow from a natural language description. "
"First generates a preview, then saves to library if save=true."
)
@property
def requires_auth(self) -> bool:
return True
@property
def parameters(self) -> dict[str, Any]:
return {
"type": "object",
"properties": {
"description": {
"type": "string",
"description": (
"Natural language description of what the agent should do. "
"Be specific about inputs, outputs, and the workflow steps."
),
},
"context": {
"type": "string",
"description": (
"Additional context or answers to previous clarifying questions. "
"Include any preferences or constraints mentioned by the user."
),
},
"save": {
"type": "boolean",
"description": (
"Whether to save the agent to the user's library. "
"Default is true. Set to false for preview only."
),
"default": True,
},
},
"required": ["description"],
}
async def _execute(
self,
user_id: str | None,
session: ChatSession,
**kwargs,
) -> ToolResponseBase:
"""Execute the create_agent tool.
Flow:
1. Decompose the description into steps (may return clarifying questions)
2. Generate agent JSON from the steps
3. Apply fixes to correct common LLM errors
4. Preview or save based on the save parameter
"""
description = kwargs.get("description", "").strip()
context = kwargs.get("context", "")
save = kwargs.get("save", True)
session_id = session.session_id if session else None
if not description:
return ErrorResponse(
message="Please provide a description of what the agent should do.",
error="Missing description parameter",
session_id=session_id,
)
# Step 1: Decompose goal into steps
try:
decomposition_result = await decompose_goal(description, context)
except ValueError as e:
# Handle missing API key or configuration errors
return ErrorResponse(
message=f"Agent generation is not configured: {str(e)}",
error="configuration_error",
session_id=session_id,
)
if decomposition_result is None:
return ErrorResponse(
message="Failed to analyze the goal. Please try rephrasing.",
error="Decomposition failed",
session_id=session_id,
)
# Check if LLM returned clarifying questions
if decomposition_result.get("type") == "clarifying_questions":
questions = decomposition_result.get("questions", [])
return ClarificationNeededResponse(
message=(
"I need some more information to create this agent. "
"Please answer the following questions:"
),
questions=[
ClarifyingQuestion(
question=q.get("question", ""),
keyword=q.get("keyword", ""),
example=q.get("example"),
)
for q in questions
],
session_id=session_id,
)
# Check for unachievable/vague goals
if decomposition_result.get("type") == "unachievable_goal":
suggested = decomposition_result.get("suggested_goal", "")
reason = decomposition_result.get("reason", "")
return ErrorResponse(
message=(
f"This goal cannot be accomplished with the available blocks. "
f"{reason} "
f"Suggestion: {suggested}"
),
error="unachievable_goal",
details={"suggested_goal": suggested, "reason": reason},
session_id=session_id,
)
if decomposition_result.get("type") == "vague_goal":
suggested = decomposition_result.get("suggested_goal", "")
return ErrorResponse(
message=(
f"The goal is too vague to create a specific workflow. "
f"Suggestion: {suggested}"
),
error="vague_goal",
details={"suggested_goal": suggested},
session_id=session_id,
)
# Step 2: Generate agent JSON with retry on validation failure
blocks_info = get_blocks_info()
agent_json = None
validation_errors = None
for attempt in range(MAX_GENERATION_RETRIES + 1):
# Generate agent (include validation errors from previous attempt)
if attempt == 0:
agent_json = await generate_agent(decomposition_result)
else:
# Retry with validation error feedback
logger.info(
f"Retry {attempt}/{MAX_GENERATION_RETRIES} with validation feedback"
)
retry_instructions = {
**decomposition_result,
"previous_errors": validation_errors,
"retry_instructions": (
"The previous generation had validation errors. "
"Please fix these issues in the new generation:\n"
f"{validation_errors}"
),
}
agent_json = await generate_agent(retry_instructions)
if agent_json is None:
if attempt == MAX_GENERATION_RETRIES:
return ErrorResponse(
message="Failed to generate the agent. Please try again.",
error="Generation failed",
session_id=session_id,
)
continue
# Step 3: Apply fixes to correct common errors
agent_json = apply_all_fixes(agent_json, blocks_info)
# Step 4: Validate the agent
is_valid, validation_errors = validate_agent(agent_json, blocks_info)
if is_valid:
logger.info(f"Agent generated successfully on attempt {attempt + 1}")
break
logger.warning(
f"Validation failed on attempt {attempt + 1}: {validation_errors}"
)
if attempt == MAX_GENERATION_RETRIES:
# Return error with validation details
return ErrorResponse(
message=(
f"Generated agent has validation errors after {MAX_GENERATION_RETRIES + 1} attempts. "
f"Please try rephrasing your request or simplify the workflow."
),
error="validation_failed",
details={"validation_errors": validation_errors},
session_id=session_id,
)
agent_name = agent_json.get("name", "Generated Agent")
agent_description = agent_json.get("description", "")
node_count = len(agent_json.get("nodes", []))
link_count = len(agent_json.get("links", []))
# Step 4: Preview or save
if not save:
return AgentPreviewResponse(
message=(
f"I've generated an agent called '{agent_name}' with {node_count} blocks. "
f"Review it and call create_agent with save=true to save it to your library."
),
agent_json=agent_json,
agent_name=agent_name,
description=agent_description,
node_count=node_count,
link_count=link_count,
session_id=session_id,
)
# Save to library
if not user_id:
return ErrorResponse(
message="You must be logged in to save agents.",
error="auth_required",
session_id=session_id,
)
try:
created_graph, library_agent = await save_agent_to_library(
agent_json, user_id
)
return AgentSavedResponse(
message=f"Agent '{created_graph.name}' has been saved to your library!",
agent_id=created_graph.id,
agent_name=created_graph.name,
library_agent_id=library_agent.id,
library_agent_link=f"/library/{library_agent.id}",
agent_page_link=f"/build?flowID={created_graph.id}",
session_id=session_id,
)
except Exception as e:
return ErrorResponse(
message=f"Failed to save the agent: {str(e)}",
error="save_failed",
details={"exception": str(e)},
session_id=session_id,
)

View File

@@ -0,0 +1,294 @@
"""EditAgentTool - Edits existing agents using natural language."""
import logging
from typing import Any
from backend.api.features.chat.model import ChatSession
from .agent_generator import (
apply_agent_patch,
apply_all_fixes,
generate_agent_patch,
get_agent_as_json,
get_blocks_info,
save_agent_to_library,
validate_agent,
)
from .base import BaseTool
from .models import (
AgentPreviewResponse,
AgentSavedResponse,
ClarificationNeededResponse,
ClarifyingQuestion,
ErrorResponse,
ToolResponseBase,
)
logger = logging.getLogger(__name__)
# Maximum retries for patch generation with validation feedback
MAX_GENERATION_RETRIES = 2
class EditAgentTool(BaseTool):
"""Tool for editing existing agents using natural language."""
@property
def name(self) -> str:
return "edit_agent"
@property
def description(self) -> str:
return (
"Edit an existing agent from the user's library using natural language. "
"Generates a patch to update the agent while preserving unchanged parts."
)
@property
def requires_auth(self) -> bool:
return True
@property
def parameters(self) -> dict[str, Any]:
return {
"type": "object",
"properties": {
"agent_id": {
"type": "string",
"description": (
"The ID of the agent to edit. "
"Can be a graph ID or library agent ID."
),
},
"changes": {
"type": "string",
"description": (
"Natural language description of what changes to make. "
"Be specific about what to add, remove, or modify."
),
},
"context": {
"type": "string",
"description": (
"Additional context or answers to previous clarifying questions."
),
},
"save": {
"type": "boolean",
"description": (
"Whether to save the changes. "
"Default is true. Set to false for preview only."
),
"default": True,
},
},
"required": ["agent_id", "changes"],
}
async def _execute(
self,
user_id: str | None,
session: ChatSession,
**kwargs,
) -> ToolResponseBase:
"""Execute the edit_agent tool.
Flow:
1. Fetch the current agent
2. Generate a patch based on the requested changes
3. Apply the patch to create an updated agent
4. Preview or save based on the save parameter
"""
agent_id = kwargs.get("agent_id", "").strip()
changes = kwargs.get("changes", "").strip()
context = kwargs.get("context", "")
save = kwargs.get("save", True)
session_id = session.session_id if session else None
if not agent_id:
return ErrorResponse(
message="Please provide the agent ID to edit.",
error="Missing agent_id parameter",
session_id=session_id,
)
if not changes:
return ErrorResponse(
message="Please describe what changes you want to make.",
error="Missing changes parameter",
session_id=session_id,
)
# Step 1: Fetch current agent
current_agent = await get_agent_as_json(agent_id, user_id)
if current_agent is None:
return ErrorResponse(
message=f"Could not find agent with ID '{agent_id}' in your library.",
error="agent_not_found",
session_id=session_id,
)
# Build the update request with context
update_request = changes
if context:
update_request = f"{changes}\n\nAdditional context:\n{context}"
# Step 2: Generate patch with retry on validation failure
blocks_info = get_blocks_info()
updated_agent = None
validation_errors = None
intent = "Applied requested changes"
for attempt in range(MAX_GENERATION_RETRIES + 1):
# Generate patch (include validation errors from previous attempt)
try:
if attempt == 0:
patch_result = await generate_agent_patch(
update_request, current_agent
)
else:
# Retry with validation error feedback
logger.info(
f"Retry {attempt}/{MAX_GENERATION_RETRIES} with validation feedback"
)
retry_request = (
f"{update_request}\n\n"
f"IMPORTANT: The previous edit had validation errors. "
f"Please fix these issues:\n{validation_errors}"
)
patch_result = await generate_agent_patch(
retry_request, current_agent
)
except ValueError as e:
# Handle missing API key or configuration errors
return ErrorResponse(
message=f"Agent generation is not configured: {str(e)}",
error="configuration_error",
session_id=session_id,
)
if patch_result is None:
if attempt == MAX_GENERATION_RETRIES:
return ErrorResponse(
message="Failed to generate changes. Please try rephrasing.",
error="Patch generation failed",
session_id=session_id,
)
continue
# Check if LLM returned clarifying questions
if patch_result.get("type") == "clarifying_questions":
questions = patch_result.get("questions", [])
return ClarificationNeededResponse(
message=(
"I need some more information about the changes. "
"Please answer the following questions:"
),
questions=[
ClarifyingQuestion(
question=q.get("question", ""),
keyword=q.get("keyword", ""),
example=q.get("example"),
)
for q in questions
],
session_id=session_id,
)
# Step 3: Apply patch and fixes
try:
updated_agent = apply_agent_patch(current_agent, patch_result)
updated_agent = apply_all_fixes(updated_agent, blocks_info)
except Exception as e:
if attempt == MAX_GENERATION_RETRIES:
return ErrorResponse(
message=f"Failed to apply changes: {str(e)}",
error="patch_apply_failed",
details={"exception": str(e)},
session_id=session_id,
)
validation_errors = str(e)
continue
# Step 4: Validate the updated agent
is_valid, validation_errors = validate_agent(updated_agent, blocks_info)
if is_valid:
logger.info(f"Agent edited successfully on attempt {attempt + 1}")
intent = patch_result.get("intent", "Applied requested changes")
break
logger.warning(
f"Validation failed on attempt {attempt + 1}: {validation_errors}"
)
if attempt == MAX_GENERATION_RETRIES:
# Return error with validation details
return ErrorResponse(
message=(
f"Updated agent has validation errors after "
f"{MAX_GENERATION_RETRIES + 1} attempts. "
f"Please try rephrasing your request or simplify the changes."
),
error="validation_failed",
details={"validation_errors": validation_errors},
session_id=session_id,
)
# At this point, updated_agent is guaranteed to be set (we return on all failure paths)
assert updated_agent is not None
agent_name = updated_agent.get("name", "Updated Agent")
agent_description = updated_agent.get("description", "")
node_count = len(updated_agent.get("nodes", []))
link_count = len(updated_agent.get("links", []))
# Step 5: Preview or save
if not save:
return AgentPreviewResponse(
message=(
f"I've updated the agent. Changes: {intent}. "
f"The agent now has {node_count} blocks. "
f"Review it and call edit_agent with save=true to save the changes."
),
agent_json=updated_agent,
agent_name=agent_name,
description=agent_description,
node_count=node_count,
link_count=link_count,
session_id=session_id,
)
# Save to library (creates a new version)
if not user_id:
return ErrorResponse(
message="You must be logged in to save agents.",
error="auth_required",
session_id=session_id,
)
try:
created_graph, library_agent = await save_agent_to_library(
updated_agent, user_id, is_update=True
)
return AgentSavedResponse(
message=(
f"Updated agent '{created_graph.name}' has been saved to your library! "
f"Changes: {intent}"
),
agent_id=created_graph.id,
agent_name=created_graph.name,
library_agent_id=library_agent.id,
library_agent_link=f"/library/{library_agent.id}",
agent_page_link=f"/build?flowID={created_graph.id}",
session_id=session_id,
)
except Exception as e:
return ErrorResponse(
message=f"Failed to save the updated agent: {str(e)}",
error="save_failed",
details={"exception": str(e)},
session_id=session_id,
)

View File

@@ -0,0 +1,157 @@
"""Tool for searching agents in the user's library."""
import logging
from typing import Any
from backend.api.features.chat.model import ChatSession
from backend.api.features.library import db as library_db
from backend.util.exceptions import DatabaseError
from .base import BaseTool
from .models import (
AgentCarouselResponse,
AgentInfo,
ErrorResponse,
NoResultsResponse,
ToolResponseBase,
)
logger = logging.getLogger(__name__)
class FindLibraryAgentTool(BaseTool):
"""Tool for searching agents in the user's library."""
@property
def name(self) -> str:
return "find_library_agent"
@property
def description(self) -> str:
return (
"Search for agents in the user's library. Use this to find agents "
"the user has already added to their library, including agents they "
"created or added from the marketplace."
)
@property
def parameters(self) -> dict[str, Any]:
return {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": (
"Search query to find agents by name or description. "
"Use keywords for best results."
),
},
},
"required": ["query"],
}
@property
def requires_auth(self) -> bool:
return True
async def _execute(
self,
user_id: str | None,
session: ChatSession,
**kwargs,
) -> ToolResponseBase:
"""Search for agents in the user's library.
Args:
user_id: User ID (required)
session: Chat session
query: Search query
Returns:
AgentCarouselResponse: List of agents found in the library
NoResultsResponse: No agents found
ErrorResponse: Error message
"""
query = kwargs.get("query", "").strip()
session_id = session.session_id
if not query:
return ErrorResponse(
message="Please provide a search query",
session_id=session_id,
)
if not user_id:
return ErrorResponse(
message="User authentication required to search library",
session_id=session_id,
)
agents = []
try:
logger.info(f"Searching user library for: {query}")
library_results = await library_db.list_library_agents(
user_id=user_id,
search_term=query,
page_size=10,
)
logger.info(
f"Find library agents tool found {len(library_results.agents)} agents"
)
for agent in library_results.agents:
agents.append(
AgentInfo(
id=agent.id,
name=agent.name,
description=agent.description or "",
source="library",
in_library=True,
creator=agent.creator_name,
status=agent.status.value,
can_access_graph=agent.can_access_graph,
has_external_trigger=agent.has_external_trigger,
new_output=agent.new_output,
graph_id=agent.graph_id,
),
)
except DatabaseError as e:
logger.error(f"Error searching library agents: {e}", exc_info=True)
return ErrorResponse(
message="Failed to search library. Please try again.",
error=str(e),
session_id=session_id,
)
if not agents:
return NoResultsResponse(
message=(
f"No agents found matching '{query}' in your library. "
"Try different keywords or use find_agent to search the marketplace."
),
session_id=session_id,
suggestions=[
"Try more general terms",
"Use find_agent to search the marketplace",
"Check your library at /library",
],
)
title = (
f"Found {len(agents)} agent{'s' if len(agents) != 1 else ''} "
f"in your library for '{query}'"
)
return AgentCarouselResponse(
message=(
"Found agents in the user's library. You can provide a link to "
"view an agent at: /library/agents/{agent_id}. "
"Use agent_output to get execution results, or run_agent to execute."
),
title=title,
agents=agents,
count=len(agents),
session_id=session_id,
)

View File

@@ -1,5 +1,6 @@
"""Pydantic models for tool responses."""
from datetime import datetime
from enum import Enum
from typing import Any
@@ -19,6 +20,15 @@ class ResponseType(str, Enum):
ERROR = "error"
NO_RESULTS = "no_results"
SUCCESS = "success"
DOC_SEARCH_RESULTS = "doc_search_results"
AGENT_OUTPUT = "agent_output"
BLOCK_LIST = "block_list"
BLOCK_OUTPUT = "block_output"
UNDERSTANDING_UPDATED = "understanding_updated"
# Agent generation responses
AGENT_PREVIEW = "agent_preview"
AGENT_SAVED = "agent_saved"
CLARIFICATION_NEEDED = "clarification_needed"
# Base response model
@@ -173,3 +183,128 @@ class ErrorResponse(ToolResponseBase):
type: ResponseType = ResponseType.ERROR
error: str | None = None
details: dict[str, Any] | None = None
# Documentation search models
class DocSearchResult(BaseModel):
"""A single documentation search result."""
title: str
path: str
section: str
snippet: str # Short excerpt for UI display
content: str # Full text content for LLM to read and understand
score: float
doc_url: str | None = None
class DocSearchResultsResponse(ToolResponseBase):
"""Response for search_docs tool."""
type: ResponseType = ResponseType.DOC_SEARCH_RESULTS
results: list[DocSearchResult]
count: int
query: str
# Agent output models
class ExecutionOutputInfo(BaseModel):
"""Summary of a single execution's outputs."""
execution_id: str
status: str
started_at: datetime | None = None
ended_at: datetime | None = None
outputs: dict[str, list[Any]]
inputs_summary: dict[str, Any] | None = None
class AgentOutputResponse(ToolResponseBase):
"""Response for agent_output tool."""
type: ResponseType = ResponseType.AGENT_OUTPUT
agent_name: str
agent_id: str
library_agent_id: str | None = None
library_agent_link: str | None = None
execution: ExecutionOutputInfo | None = None
available_executions: list[dict[str, Any]] | None = None
total_executions: int = 0
# Block models
class BlockInfoSummary(BaseModel):
"""Summary of a block for search results."""
id: str
name: str
description: str
categories: list[str]
input_schema: dict[str, Any]
output_schema: dict[str, Any]
class BlockListResponse(ToolResponseBase):
"""Response for find_block tool."""
type: ResponseType = ResponseType.BLOCK_LIST
blocks: list[BlockInfoSummary]
count: int
query: str
class BlockOutputResponse(ToolResponseBase):
"""Response for run_block tool."""
type: ResponseType = ResponseType.BLOCK_OUTPUT
block_id: str
block_name: str
outputs: dict[str, list[Any]]
success: bool = True
# Business understanding models
class UnderstandingUpdatedResponse(ToolResponseBase):
"""Response for add_understanding tool."""
type: ResponseType = ResponseType.UNDERSTANDING_UPDATED
updated_fields: list[str] = Field(default_factory=list)
current_understanding: dict[str, Any] = Field(default_factory=dict)
# Agent generation models
class ClarifyingQuestion(BaseModel):
"""A question that needs user clarification."""
question: str
keyword: str
example: str | None = None
class AgentPreviewResponse(ToolResponseBase):
"""Response for previewing a generated agent before saving."""
type: ResponseType = ResponseType.AGENT_PREVIEW
agent_json: dict[str, Any]
agent_name: str
description: str
node_count: int
link_count: int = 0
class AgentSavedResponse(ToolResponseBase):
"""Response when an agent is saved to the library."""
type: ResponseType = ResponseType.AGENT_SAVED
agent_id: str
agent_name: str
library_agent_id: str
library_agent_link: str
agent_page_link: str # Link to the agent builder/editor page
class ClarificationNeededResponse(ToolResponseBase):
"""Response when the LLM needs more information from the user."""
type: ResponseType = ResponseType.CLARIFICATION_NEEDED
questions: list[ClarifyingQuestion] = Field(default_factory=list)

View File

@@ -7,6 +7,7 @@ from pydantic import BaseModel, Field, field_validator
from backend.api.features.chat.config import ChatConfig
from backend.api.features.chat.model import ChatSession
from backend.api.features.library import db as library_db
from backend.data.graph import GraphModel
from backend.data.model import CredentialsMetaInput
from backend.data.user import get_user_by_id
@@ -57,6 +58,7 @@ class RunAgentInput(BaseModel):
"""Input parameters for the run_agent tool."""
username_agent_slug: str = ""
library_agent_id: str = ""
inputs: dict[str, Any] = Field(default_factory=dict)
use_defaults: bool = False
schedule_name: str = ""
@@ -64,7 +66,12 @@ class RunAgentInput(BaseModel):
timezone: str = "UTC"
@field_validator(
"username_agent_slug", "schedule_name", "cron", "timezone", mode="before"
"username_agent_slug",
"library_agent_id",
"schedule_name",
"cron",
"timezone",
mode="before",
)
@classmethod
def strip_strings(cls, v: Any) -> Any:
@@ -90,7 +97,7 @@ class RunAgentTool(BaseTool):
@property
def description(self) -> str:
return """Run or schedule an agent from the marketplace.
return """Run or schedule an agent from the marketplace or user's library.
The tool automatically handles the setup flow:
- Returns missing inputs if required fields are not provided
@@ -98,6 +105,10 @@ class RunAgentTool(BaseTool):
- Executes immediately if all requirements are met
- Schedules execution if cron expression is provided
Identify the agent using either:
- username_agent_slug: Marketplace format 'username/agent-name'
- library_agent_id: ID of an agent in the user's library
For scheduled execution, provide: schedule_name, cron, and optionally timezone."""
@property
@@ -109,6 +120,10 @@ class RunAgentTool(BaseTool):
"type": "string",
"description": "Agent identifier in format 'username/agent-name'",
},
"library_agent_id": {
"type": "string",
"description": "Library agent ID from user's library",
},
"inputs": {
"type": "object",
"description": "Input values for the agent",
@@ -131,7 +146,7 @@ class RunAgentTool(BaseTool):
"description": "IANA timezone for schedule (default: UTC)",
},
},
"required": ["username_agent_slug"],
"required": [],
}
@property
@@ -149,10 +164,16 @@ class RunAgentTool(BaseTool):
params = RunAgentInput(**kwargs)
session_id = session.session_id
# Validate agent slug format
if not params.username_agent_slug or "/" not in params.username_agent_slug:
# Validate at least one identifier is provided
has_slug = params.username_agent_slug and "/" in params.username_agent_slug
has_library_id = bool(params.library_agent_id)
if not has_slug and not has_library_id:
return ErrorResponse(
message="Please provide an agent slug in format 'username/agent-name'",
message=(
"Please provide either a username_agent_slug "
"(format 'username/agent-name') or a library_agent_id"
),
session_id=session_id,
)
@@ -167,13 +188,41 @@ class RunAgentTool(BaseTool):
is_schedule = bool(params.schedule_name or params.cron)
try:
# Step 1: Fetch agent details (always happens first)
username, agent_name = params.username_agent_slug.split("/", 1)
graph, store_agent = await fetch_graph_from_store_slug(username, agent_name)
# Step 1: Fetch agent details
graph: GraphModel | None = None
library_agent = None
# Priority: library_agent_id if provided
if has_library_id:
library_agent = await library_db.get_library_agent(
params.library_agent_id, user_id
)
if not library_agent:
return ErrorResponse(
message=f"Library agent '{params.library_agent_id}' not found",
session_id=session_id,
)
# Get the graph from the library agent
from backend.data.graph import get_graph
graph = await get_graph(
library_agent.graph_id,
library_agent.graph_version,
user_id=user_id,
)
else:
# Fetch from marketplace slug
username, agent_name = params.username_agent_slug.split("/", 1)
graph, _ = await fetch_graph_from_store_slug(username, agent_name)
if not graph:
identifier = (
params.library_agent_id
if has_library_id
else params.username_agent_slug
)
return ErrorResponse(
message=f"Agent '{params.username_agent_slug}' not found in marketplace",
message=f"Agent '{identifier}' not found",
session_id=session_id,
)

View File

@@ -0,0 +1,78 @@
-- CreateTable
CREATE TABLE "UserBusinessUnderstanding" (
"id" TEXT NOT NULL,
"createdAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
"updatedAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
"userId" TEXT NOT NULL,
"userName" TEXT,
"jobTitle" TEXT,
"businessName" TEXT,
"industry" TEXT,
"businessSize" TEXT,
"userRole" TEXT,
"keyWorkflows" JSONB,
"dailyActivities" JSONB,
"painPoints" JSONB,
"bottlenecks" JSONB,
"manualTasks" JSONB,
"automationGoals" JSONB,
"currentSoftware" JSONB,
"existingAutomation" JSONB,
"additionalNotes" TEXT,
CONSTRAINT "UserBusinessUnderstanding_pkey" PRIMARY KEY ("id")
);
-- CreateTable
CREATE TABLE "ChatSession" (
"id" TEXT NOT NULL,
"createdAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
"updatedAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
"userId" TEXT,
"title" TEXT,
"credentials" JSONB NOT NULL DEFAULT '{}',
"successfulAgentRuns" JSONB NOT NULL DEFAULT '{}',
"successfulAgentSchedules" JSONB NOT NULL DEFAULT '{}',
"totalPromptTokens" INTEGER NOT NULL DEFAULT 0,
"totalCompletionTokens" INTEGER NOT NULL DEFAULT 0,
CONSTRAINT "ChatSession_pkey" PRIMARY KEY ("id")
);
-- CreateTable
CREATE TABLE "ChatMessage" (
"id" TEXT NOT NULL,
"createdAt" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
"sessionId" TEXT NOT NULL,
"role" TEXT NOT NULL,
"content" TEXT,
"name" TEXT,
"toolCallId" TEXT,
"refusal" TEXT,
"toolCalls" JSONB,
"functionCall" JSONB,
"sequence" INTEGER NOT NULL,
CONSTRAINT "ChatMessage_pkey" PRIMARY KEY ("id")
);
-- CreateIndex
CREATE UNIQUE INDEX "UserBusinessUnderstanding_userId_key" ON "UserBusinessUnderstanding"("userId");
-- CreateIndex
CREATE INDEX "UserBusinessUnderstanding_userId_idx" ON "UserBusinessUnderstanding"("userId");
-- CreateIndex
CREATE INDEX "ChatSession_userId_updatedAt_idx" ON "ChatSession"("userId", "updatedAt");
-- CreateIndex
CREATE INDEX "ChatMessage_sessionId_sequence_idx" ON "ChatMessage"("sessionId", "sequence");
-- CreateIndex
CREATE UNIQUE INDEX "ChatMessage_sessionId_sequence_key" ON "ChatMessage"("sessionId", "sequence");
-- AddForeignKey
ALTER TABLE "UserBusinessUnderstanding" ADD CONSTRAINT "UserBusinessUnderstanding_userId_fkey" FOREIGN KEY ("userId") REFERENCES "User"("id") ON DELETE CASCADE ON UPDATE CASCADE;
-- AddForeignKey
ALTER TABLE "ChatMessage" ADD CONSTRAINT "ChatMessage_sessionId_fkey" FOREIGN KEY ("sessionId") REFERENCES "ChatSession"("id") ON DELETE CASCADE ON UPDATE CASCADE;