Spaces:
Running
Running
from fastapi import FastAPI, HTTPException, Query as QueryParam | |
from pydantic import BaseModel, Field | |
from langchain_openai import ChatOpenAI, OpenAIEmbeddings | |
from qdrant_client import QdrantClient | |
from langchain.agents import Tool, AgentExecutor, create_openai_tools_agent | |
from langchain.prompts import ChatPromptTemplate, MessagesPlaceholder | |
from langchain.tools import BaseTool | |
from langchain.memory import ConversationBufferMemory | |
from langchain_core.messages import HumanMessage, AIMessage | |
from typing import Type, Optional, List, Dict, Any | |
import os | |
import warnings | |
import base64 | |
import requests | |
from dotenv import load_dotenv | |
from datetime import datetime | |
import json | |
# Suppress warnings | |
warnings.filterwarnings("ignore", message="Qdrant client version.*is incompatible.*") | |
load_dotenv() | |
app = FastAPI(title="AI Agent with Document Search, Project Management and Session Memory") | |
# Environment variables | |
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") | |
QDRANT_COLLECTION_NAME = os.getenv("QDRANT_COLLECTION_NAME", "vatsav_test_1") | |
QDRANT_HOST = os.getenv("QDRANT_HOST", "127.0.0.1") | |
QDRANT_PORT = int(os.getenv("QDRANT_PORT", 6333)) | |
# Initialize models | |
embedding_model = OpenAIEmbeddings( | |
model="text-embedding-3-large", | |
openai_api_key=OPENAI_API_KEY, | |
) | |
qdrant_client = QdrantClient(host=QDRANT_HOST, port=QDRANT_PORT) | |
llm = ChatOpenAI(model="gpt-4o", temperature=0, openai_api_key=OPENAI_API_KEY) | |
# === USER SESSION MANAGEMENT === | |
# In-memory storage for user sessions (in production, use Redis or database) | |
user_memories: Dict[int, ConversationBufferMemory] = {} | |
user_chat_history: Dict[int, List[Dict]] = {} | |
def get_or_create_user_memory(user_login_id: int) -> ConversationBufferMemory: | |
"""Get or create a conversation memory for a user""" | |
if user_login_id not in user_memories: | |
user_memories[user_login_id] = ConversationBufferMemory( | |
memory_key="chat_history", | |
return_messages=True | |
) | |
user_chat_history[user_login_id] = [] | |
return user_memories[user_login_id] | |
def add_to_chat_history(user_login_id: int, user_message: str, ai_response: str): | |
"""Add messages to user chat history""" | |
if user_login_id not in user_chat_history: | |
user_chat_history[user_login_id] = [] | |
timestamp = datetime.now().isoformat() | |
# Add user message | |
user_chat_history[user_login_id].append({ | |
"id": len(user_chat_history[user_login_id]) + 1, | |
"type": "user", | |
"message": user_message, | |
"timestamp": timestamp | |
}) | |
# Add AI response | |
user_chat_history[user_login_id].append({ | |
"id": len(user_chat_history[user_login_id]) + 1, | |
"type": "assistant", | |
"message": ai_response, | |
"timestamp": timestamp | |
}) | |
# === INPUT SCHEMAS === | |
class Query(BaseModel): | |
message: str | |
class ProjectRequest(BaseModel): | |
userLoginId: int | |
orgId: int | |
auth_token: str | |
class AgentQuery(BaseModel): | |
message: str | |
userLoginId: int # Now required for user-based memory | |
orgId: Optional[int] = None | |
auth_token: Optional[str] = None | |
class ChatHistoryResponse(BaseModel): | |
user_login_id: int | |
total_messages: int | |
messages: List[Dict[str, Any]] | |
# === UTILITY FUNCTIONS === | |
def get_encoded_auth_token(user: int, token: str) -> str: | |
auth_string = f"{user}:{token}" | |
return base64.b64encode(auth_string.encode("utf-8")).decode("utf-8") | |
def fetch_user_projects(userLoginId: int, orgId: int, auth_token: str): | |
url = "https://japidemo.dev.ingenspark.com/fetchUserProjects" | |
payload = { | |
"userLoginId": userLoginId, | |
"orgId": orgId | |
} | |
headers = { | |
'accept': 'application/json, text/plain, */*', | |
'authorization': f'Basic {auth_token}', | |
'content-type': 'application/json; charset=UTF-8' | |
} | |
print("auth_token", auth_token) | |
try: | |
response = requests.post(url, headers=headers, json=payload) | |
response.raise_for_status() | |
return response.json() | |
except requests.exceptions.RequestException as e: | |
raise HTTPException(status_code=response.status_code if 'response' in locals() else 500, | |
detail=str(e)) | |
def format_project_response(data: dict) -> str: | |
my_projects = data.get("data", {}).get("Myprojects", []) | |
other_projects = data.get("data", {}).get("Otherprojects", []) | |
all_projects = [] | |
for project in my_projects: | |
all_projects.append({ | |
"type": "Your Project", | |
"projectNm": project["projectNm"], | |
"projectId": project["projectId"], | |
"created_dttm": project["created_dttm"].split('.')[0], | |
"description": project["description"], | |
"categoryName": project["categoryName"] | |
}) | |
for project in other_projects: | |
all_projects.append({ | |
"type": "Other Project", | |
"projectNm": project["projectNm"], | |
"projectId": project["projectId"], | |
"created_dttm": project["created_dttm"].split('.')[0], | |
"description": project["description"], | |
"categoryName": project["categoryName"] | |
}) | |
if not all_projects: | |
return "β No projects found." | |
# Build the formatted string | |
result = [f"β You have access to {len(all_projects)} project(s):\n"] | |
for i, project in enumerate(all_projects, 1): | |
result.append(f"{i}. Project Name: {project['projectNm']} ({project['type']})") | |
result.append(f" Project ID: {project['projectId']}") | |
result.append(f" Created On: {project['created_dttm']}") | |
result.append(f" Description: {project['description']}") | |
result.append(f" Category: {project['categoryName']}\n") | |
return "\n".join(result) | |
# === TOOL FUNCTIONS === | |
def search_documents(query: str) -> str: | |
"""Search through ingested documents and get relevant information. | |
Args: | |
query: The search query or message about the documents | |
Returns: | |
Relevant information from the documents with sources | |
""" | |
try: | |
# Generate embedding for the query | |
query_vector = embedding_model.embed_query(query) | |
# Search in Qdrant | |
search_result = qdrant_client.search( | |
collection_name=QDRANT_COLLECTION_NAME, | |
query_vector=query_vector, | |
limit=5, | |
) | |
if not search_result: | |
return "No relevant information found in the knowledge base." | |
# Convert results to text content | |
context_texts = [] | |
sources = [] | |
for hit in search_result: | |
context_texts.append(hit.payload["text"]) | |
sources.append(hit.payload.get("source", "unknown")) | |
# Create a simple prompt for answering based on context | |
context = "\n\n".join(context_texts) | |
unique_sources = list(set(sources)) | |
# Use the LLM directly to answer the message based on context | |
prompt = f"""Based on the following context, answer the message: {query} | |
Context: | |
{context} | |
Please provide a comprehensive answer based on the context above. If the context doesn't contain enough information to answer the message, say so clearly.""" | |
response = llm.invoke(prompt) | |
return f"{response.content}\n\nSources: {', '.join(unique_sources)}" | |
except Exception as e: | |
return f"Error searching documents: {str(e)}" | |
# Global variables to store auth context (for tool functions) | |
_current_user_id = None | |
_current_org_id = None | |
_current_auth_token = None | |
def get_user_projects(userLoginId: str) -> str: | |
"""Get list of projects for a user. Requires userLoginId as input. | |
Args: | |
userLoginId: The user login ID to fetch projects for (format: 'userLoginId:orgId' or just 'userLoginId') | |
Returns: | |
Formatted list of user projects | |
""" | |
try: | |
# Use global auth context if available | |
if _current_auth_token and _current_user_id: | |
user_id = _current_user_id | |
org_id = _current_org_id or 1 | |
auth_token = _current_auth_token | |
else: | |
# Parse userLoginId (can be "25" or "25:1" format) - fallback | |
if ":" in userLoginId: | |
user_id, org_id = userLoginId.split(":", 1) | |
user_id = int(user_id) | |
org_id = int(org_id) | |
else: | |
user_id = int(userLoginId) | |
org_id = 1 # Default org ID | |
return "β Authentication token required. Please provide auth_token in your request." | |
# Encode auth token using the actual user ID and provided token | |
encoded_token = get_encoded_auth_token(user_id, auth_token) | |
# Fetch projects | |
data = fetch_user_projects(user_id, org_id, encoded_token) | |
print("Fetched data:", data) # Debugging line | |
# Format and return the project list | |
formatted = format_project_response(data) | |
return formatted | |
except ValueError: | |
return "β Invalid userLoginId format. Please provide a valid number or 'userLoginId:orgId' format." | |
except Exception as e: | |
return f"β Error fetching projects: {str(e)}" | |
# === CREATE TOOLS === | |
document_search_tool = Tool( | |
name="document_search", | |
description="""Use this tool to search through ingested documents and get relevant information from the knowledge base. | |
Perfect for answering messages about uploaded documents, manuals, or any content that was previously stored. | |
Input should be a search query or message about the documents.""", | |
func=search_documents | |
) | |
project_list_tool = Tool( | |
name="get_user_projects", | |
description="""Use this tool to get the list of projects for a user. | |
Perfect for when users ask about their projects, want to see available projects, or need project information. | |
Input should be the userLoginId (e.g., '25') or in format 'userLoginId:orgId' (e.g., '25:1'). | |
Note: This tool requires authentication context to be set.""", | |
func=get_user_projects | |
) | |
# === AGENT SETUP === | |
def create_agent_with_memory(memory: ConversationBufferMemory): | |
"""Create agent with session memory""" | |
agent_prompt = ChatPromptTemplate.from_messages([ | |
("system", """You are a helpful AI assistant with access to multiple tools and conversation memory: | |
1. **Document Search**: Search through uploaded documents and knowledge base | |
2. **Project Management**: Get list of user projects and project information | |
Your capabilities: | |
- Answer messages about documents using the document search tool | |
- Help users find their projects and project information | |
- Remember previous conversations in this session | |
- Provide general assistance and information | |
- Use appropriate tools based on user queries | |
Guidelines: | |
- Use the document search tool when users ask about specific content, documentation, or information that might be in uploaded files | |
- Use the project tool when users ask about projects, want to see their projects, or need project-related information | |
- Reference previous conversation context when relevant | |
- If users mention a userLoginId or ask about projects, use the project tool | |
- Be clear about which tool you're using and what information you're providing | |
- If you're unsure which tool to use, you can ask for clarification | |
- Provide helpful, accurate, and well-formatted responses | |
Remember: Always use the most appropriate tool based on the user's message and conversation context to provide the best possible answer."""), | |
MessagesPlaceholder(variable_name="chat_history"), | |
("user", "{input}"), | |
MessagesPlaceholder(variable_name="agent_scratchpad"), | |
]) | |
# Create tools list | |
tools = [document_search_tool, project_list_tool] | |
# Create the agent | |
agent = create_openai_tools_agent(llm, tools, agent_prompt) | |
# Create the agent executor with memory | |
agent_executor = AgentExecutor( | |
agent=agent, | |
tools=tools, | |
verbose=True, | |
memory=memory | |
) | |
return agent_executor | |
# === API ENDPOINTS === | |
@app.post("/bot") | |
def chat_with_agent(query: AgentQuery): | |
"""Main agent endpoint with user-based memory - handles both document search and project queries intelligently""" | |
try: | |
# Set global auth context for tools | |
global _current_user_id, _current_org_id, _current_auth_token | |
_current_user_id = query.userLoginId | |
_current_org_id = query.orgId | |
_current_auth_token = query.auth_token | |
# Get or create user memory | |
memory = get_or_create_user_memory(query.userLoginId) | |
# Create agent with memory | |
agent_executor = create_agent_with_memory(memory) | |
# Prepare the input for the agent | |
agent_input = query.message | |
# If user provided credentials, add them to the context | |
if query.userLoginId is not None: | |
agent_input += f" [UserLoginId: {query.userLoginId}" | |
if query.orgId is not None: | |
agent_input += f", OrgId: {query.orgId}" | |
agent_input += "]" | |
# Use the agent to process the query | |
result = agent_executor.invoke({"input": agent_input}) | |
# Add to chat history | |
add_to_chat_history(query.userLoginId, query.message, result["output"]) | |
# Clear auth context after use | |
_current_user_id = None | |
_current_org_id = None | |
_current_auth_token = None | |
return { | |
"message": query.message, | |
"answer": result["output"], | |
"user_login_id": query.userLoginId, | |
"agent_used": True | |
} | |
except Exception as e: | |
# Clear auth context on error | |
_current_user_id = None | |
_current_org_id = None | |
_current_auth_token = None | |
return { | |
"message": query.message, | |
"answer": f"An error occurred: {str(e)}", | |
"user_login_id": query.userLoginId, | |
"agent_used": True | |
} | |
@app.get("/get-chat-history/{user_login_id}") | |
def get_chat_history( | |
user_login_id: int, | |
n: int = QueryParam(10, description="Number of recent messages to return") | |
) -> ChatHistoryResponse: | |
"""Get chat history for a user""" | |
try: | |
if user_login_id not in user_chat_history: | |
return ChatHistoryResponse( | |
user_login_id=user_login_id, | |
total_messages=0, | |
messages=[] | |
) | |
history = user_chat_history[user_login_id] | |
# Get the last n messages (or all if less than n) | |
recent_messages = history[-n:] if len(history) > n else history | |
return ChatHistoryResponse( | |
user_login_id=user_login_id, | |
total_messages=len(history), | |
messages=recent_messages | |
) | |
except Exception as e: | |
raise HTTPException(status_code=500, detail=f"Error fetching chat history: {str(e)}") | |
@app.delete("/clear-user-history/{user_login_id}") | |
def clear_user_history(user_login_id: int): | |
"""Clear user memory and chat history""" | |
try: | |
if user_login_id in user_memories: | |
del user_memories[user_login_id] | |
if user_login_id in user_chat_history: | |
del user_chat_history[user_login_id] | |
return { | |
"message": f"User {user_login_id} chat history cleared successfully", | |
"user_login_id": user_login_id | |
} | |
except Exception as e: | |
raise HTTPException(status_code=500, detail=f"Error clearing user history: {str(e)}") | |
@app.get("/active-users") | |
def get_active_users(): | |
"""Get list of active users with chat history""" | |
return { | |
"active_users": list(user_memories.keys()), | |
"total_active_users": len(user_memories) | |
} | |
@app.post("/chat-documents") | |
def chat_documents_only(query: Query): | |
"""Direct document search without agent""" | |
try: | |
result = search_documents(query.message) | |
return { | |
"message": query.message, | |
"answer": result, | |
"tool_used": "document_search" | |
} | |
except Exception as e: | |
return { | |
"message": query.message, | |
"answer": f"An error occurred: {str(e)}", | |
"tool_used": "document_search" | |
} | |
@app.post("/list-projects") | |
def list_projects(request: ProjectRequest): | |
"""Direct project listing without agent""" | |
try: | |
# Use the provided auth token and userLoginId | |
encoded_token = get_encoded_auth_token(request.userLoginId, request.auth_token) | |
print("Encoded token:", encoded_token) | |
# Fetch projects | |
data = fetch_user_projects(request.userLoginId, request.orgId, encoded_token) | |
# Format and return the project list | |
formatted = format_project_response(data) | |
return { | |
"projects": formatted, | |
"tool_used": "project_list" | |
} | |
except Exception as e: | |
return { | |
"error": f"An error occurred: {str(e)}", | |
"tool_used": "project_list" | |
} | |
@app.get("/health") | |
def health(): | |
return { | |
"status": "ok", | |
"tools": ["document_search", "project_list"], | |
"agent": "active", | |
"user_memory_management": "enabled", | |
"active_users": len(user_memories) | |
} | |
if __name__ == "__main__": | |
import uvicorn | |
try: | |
uvicorn.run(app, host="0.0.0.0", port=8000) | |
except KeyboardInterrupt: | |
print("\nπ Server stopped gracefully") | |
except Exception as e: | |
print(f"β Server error: {e}") |