feat!: migrate to OpenAI native tool calls and fix circular deps (#fuck-gemini)
- Fix circular dependencies in agent/tools - Migrate from custom JSON to OpenAI tool calls format - Add async streaming (step_stream, complete_stream) - Simplify prompt system and remove token counting - Add 5 new API endpoints (/health, /v1/models, /api/memory/*) - Add 3 new tools (get_torrent_by_index, add_torrent_by_index, set_language) - Fix all 500 tests and add coverage config (80% threshold) - Add comprehensive docs (README, pytest guide) BREAKING: LLM interface changed, memory injection via get_memory()
This commit is contained in:
@@ -1,96 +1,219 @@
|
||||
# app.py
|
||||
"""FastAPI application for the media library agent."""
|
||||
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
import time
|
||||
import uuid
|
||||
import json
|
||||
from typing import Any, Dict
|
||||
from typing import Any
|
||||
|
||||
from fastapi import FastAPI, Request
|
||||
from fastapi import FastAPI, HTTPException
|
||||
from fastapi.responses import JSONResponse, StreamingResponse
|
||||
from pydantic import BaseModel, Field, validator
|
||||
|
||||
from agent.llm.deepseek import DeepSeekClient
|
||||
from agent.llm.ollama import OllamaClient
|
||||
from infrastructure.persistence.memory import Memory
|
||||
from agent.agent import Agent
|
||||
import os
|
||||
from agent.config import settings
|
||||
from agent.llm.deepseek import DeepSeekClient
|
||||
from agent.llm.exceptions import LLMAPIError, LLMConfigurationError
|
||||
from agent.llm.ollama import OllamaClient
|
||||
from infrastructure.persistence import get_memory, init_memory
|
||||
|
||||
logging.basicConfig(
|
||||
level=logging.INFO, format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
|
||||
)
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
app = FastAPI(
|
||||
title="LibreChat Agent Backend",
|
||||
version="0.1.0",
|
||||
title="Agent Media API",
|
||||
description="AI agent for managing a local media library",
|
||||
version="0.2.0",
|
||||
)
|
||||
|
||||
# Choose LLM based on environment variable
|
||||
# Initialize memory context at startup
|
||||
init_memory(storage_dir="memory_data")
|
||||
logger.info("Memory context initialized")
|
||||
|
||||
# Initialize LLM based on environment variable
|
||||
llm_provider = os.getenv("LLM_PROVIDER", "deepseek").lower()
|
||||
|
||||
if llm_provider == "ollama":
|
||||
print("🦙 Using Ollama LLM")
|
||||
llm = OllamaClient()
|
||||
else:
|
||||
print("🤖 Using DeepSeek LLM")
|
||||
llm = DeepSeekClient()
|
||||
try:
|
||||
if llm_provider == "ollama":
|
||||
logger.info("Using Ollama LLM")
|
||||
llm = OllamaClient()
|
||||
else:
|
||||
logger.info("Using DeepSeek LLM")
|
||||
llm = DeepSeekClient()
|
||||
except LLMConfigurationError as e:
|
||||
logger.error(f"Failed to initialize LLM: {e}")
|
||||
raise
|
||||
|
||||
memory = Memory()
|
||||
agent = Agent(llm=llm, memory=memory)
|
||||
# Initialize agent
|
||||
agent = Agent(llm=llm, max_tool_iterations=settings.max_tool_iterations)
|
||||
logger.info("Agent Media API initialized")
|
||||
|
||||
|
||||
def extract_last_user_content(messages: list[Dict[str, Any]]) -> str:
|
||||
last = ""
|
||||
# Pydantic models for request validation
|
||||
class ChatMessage(BaseModel):
|
||||
"""A single message in the conversation."""
|
||||
|
||||
role: str = Field(..., description="Role of the message sender")
|
||||
content: str | None = Field(None, description="Content of the message")
|
||||
|
||||
@validator("content")
|
||||
def content_must_not_be_empty_for_user(cls, v, values):
|
||||
"""Validate that user messages have non-empty content."""
|
||||
if values.get("role") == "user" and not v:
|
||||
raise ValueError("User messages must have non-empty content")
|
||||
return v
|
||||
|
||||
|
||||
class ChatCompletionRequest(BaseModel):
|
||||
"""Request body for chat completions."""
|
||||
|
||||
model: str = Field(default="agent-media", description="Model to use")
|
||||
messages: list[ChatMessage] = Field(..., description="List of messages")
|
||||
stream: bool = Field(default=False, description="Whether to stream the response")
|
||||
temperature: float | None = Field(default=None, ge=0.0, le=2.0)
|
||||
max_tokens: int | None = Field(default=None, gt=0)
|
||||
|
||||
@validator("messages")
|
||||
def messages_must_have_user_message(cls, v):
|
||||
"""Validate that there is at least one user message."""
|
||||
if not any(msg.role == "user" for msg in v):
|
||||
raise ValueError("At least one user message is required")
|
||||
return v
|
||||
|
||||
|
||||
def extract_last_user_content(messages: list[dict[str, Any]]) -> str:
|
||||
"""
|
||||
Extract the last user message from the conversation.
|
||||
|
||||
Args:
|
||||
messages: List of message dictionaries.
|
||||
|
||||
Returns:
|
||||
Content of the last user message, or empty string.
|
||||
"""
|
||||
for m in reversed(messages):
|
||||
if m.get("role") == "user":
|
||||
last = m.get("content") or ""
|
||||
break
|
||||
return last
|
||||
return m.get("content") or ""
|
||||
return ""
|
||||
|
||||
|
||||
@app.get("/health")
|
||||
async def health_check():
|
||||
"""Health check endpoint."""
|
||||
return {"status": "healthy", "version": "0.2.0"}
|
||||
|
||||
|
||||
@app.get("/v1/models")
|
||||
async def list_models():
|
||||
"""List available models (OpenAI-compatible endpoint)."""
|
||||
return {
|
||||
"object": "list",
|
||||
"data": [
|
||||
{
|
||||
"id": "agent-media",
|
||||
"object": "model",
|
||||
"created": int(time.time()),
|
||||
"owned_by": "local",
|
||||
}
|
||||
],
|
||||
}
|
||||
|
||||
|
||||
@app.get("/memory/state")
|
||||
async def get_memory_state():
|
||||
"""Debug endpoint to view full memory state."""
|
||||
memory = get_memory()
|
||||
return memory.get_full_state()
|
||||
|
||||
|
||||
@app.get("/memory/episodic/search-results")
|
||||
async def get_search_results():
|
||||
"""Debug endpoint to view last search results."""
|
||||
memory = get_memory()
|
||||
if memory.episodic.last_search_results:
|
||||
return {
|
||||
"status": "ok",
|
||||
"query": memory.episodic.last_search_results.get("query"),
|
||||
"type": memory.episodic.last_search_results.get("type"),
|
||||
"timestamp": memory.episodic.last_search_results.get("timestamp"),
|
||||
"result_count": len(memory.episodic.last_search_results.get("results", [])),
|
||||
"results": memory.episodic.last_search_results.get("results", []),
|
||||
}
|
||||
return {"status": "empty", "message": "No search results in episodic memory"}
|
||||
|
||||
|
||||
@app.post("/memory/clear-session")
|
||||
async def clear_session():
|
||||
"""Clear session memories (STM + Episodic)."""
|
||||
memory = get_memory()
|
||||
memory.clear_session()
|
||||
return {"status": "ok", "message": "Session memories cleared"}
|
||||
|
||||
|
||||
@app.post("/v1/chat/completions")
|
||||
async def chat_completions(request: Request):
|
||||
body = await request.json()
|
||||
model = body.get("model", "local-deepseek-agent")
|
||||
messages = body.get("messages", [])
|
||||
stream = body.get("stream", False)
|
||||
async def chat_completions(chat_request: ChatCompletionRequest):
|
||||
"""
|
||||
OpenAI-compatible chat completions endpoint.
|
||||
|
||||
user_input = extract_last_user_content(messages)
|
||||
print("Received chat completion request, stream =", stream, "input:", user_input)
|
||||
Accepts messages and returns agent response.
|
||||
Supports both streaming and non-streaming modes.
|
||||
"""
|
||||
# Convert Pydantic models to dicts for processing
|
||||
messages_dict = [msg.dict() for msg in chat_request.messages]
|
||||
|
||||
# Process user input through the agent
|
||||
answer = agent.step(user_input)
|
||||
user_input = extract_last_user_content(messages_dict)
|
||||
|
||||
logger.info(
|
||||
f"Chat request - stream={chat_request.stream}, input_length={len(user_input)}"
|
||||
)
|
||||
|
||||
try:
|
||||
answer = agent.step(user_input)
|
||||
except LLMAPIError as e:
|
||||
logger.error(f"LLM API error: {e}")
|
||||
raise HTTPException(status_code=502, detail=f"LLM API error: {e}")
|
||||
except Exception as e:
|
||||
logger.error(f"Agent error: {e}", exc_info=True)
|
||||
raise HTTPException(status_code=500, detail="Internal agent error")
|
||||
|
||||
# Ensuite = même logique de réponse (non-stream ou stream)
|
||||
created_ts = int(time.time())
|
||||
completion_id = f"chatcmpl-{uuid.uuid4().hex}"
|
||||
|
||||
if not stream:
|
||||
resp = {
|
||||
"id": completion_id,
|
||||
"object": "chat.completion",
|
||||
"created": created_ts,
|
||||
"model": model,
|
||||
"choices": [
|
||||
{
|
||||
"index": 0,
|
||||
"finish_reason": "stop",
|
||||
"message": {
|
||||
"role": "assistant",
|
||||
"content": answer or "",
|
||||
},
|
||||
}
|
||||
],
|
||||
"usage": {"prompt_tokens": 0, "completion_tokens": 0, "total_tokens": 0},
|
||||
}
|
||||
return JSONResponse(resp)
|
||||
if not chat_request.stream:
|
||||
return JSONResponse(
|
||||
{
|
||||
"id": completion_id,
|
||||
"object": "chat.completion",
|
||||
"created": created_ts,
|
||||
"model": chat_request.model,
|
||||
"choices": [
|
||||
{
|
||||
"index": 0,
|
||||
"finish_reason": "stop",
|
||||
"message": {"role": "assistant", "content": answer or ""},
|
||||
}
|
||||
],
|
||||
"usage": {
|
||||
"prompt_tokens": 0,
|
||||
"completion_tokens": 0,
|
||||
"total_tokens": 0,
|
||||
},
|
||||
}
|
||||
)
|
||||
|
||||
async def event_generator():
|
||||
chunk = {
|
||||
"id": completion_id,
|
||||
"object": "chat.completion.chunk",
|
||||
"created": created_ts,
|
||||
"model": model,
|
||||
"model": chat_request.model,
|
||||
"choices": [
|
||||
{
|
||||
"index": 0,
|
||||
"delta": {
|
||||
"role": "assistant",
|
||||
"content": answer or "",
|
||||
},
|
||||
"delta": {"role": "assistant", "content": answer or ""},
|
||||
"finish_reason": "stop",
|
||||
}
|
||||
],
|
||||
|
||||
Reference in New Issue
Block a user