Update the retrieve doc

agent_api.py

@@ -129,7 +129,7 @@ class CyberLegalAPI:
         # Initialize Resend
         resend.api_key = os.getenv("RESEND_API_KEY")
         logger.info("✅ Resend client initialized")
-
+        
         self.agent_client = CyberLegalAgent(llm=llm, tools=tools.tools_for_client,tools_facade=tools.tools_for_client_facade)
         self.agent_lawyer = CyberLegalAgent(llm=llm, tools=tools.tools_for_lawyer,tools_facade=tools.tools_for_lawyer_facade)
         self.pdf_analyzer = PDFAnalyzerAgent(llm=llm, mistral_client=mistral_client)
@@ -283,7 +283,7 @@ class CyberLegalAPI:
             logger.info(f"🤖 Calling agent.process_query with jurisdiction: {request.jurisdiction}")
             result = await agent.process_query(
                 user_query=request.message,
-                client_id=request.clientId,
+                user_id=request.userId,
                 conversation_history=conversation_history,
                 jurisdiction=request.jurisdiction,
                 system_prompt=system_prompt
@@ -415,7 +415,7 @@ class CyberLegalAPI:
         logger.info("=" * 80)
         logger.info("📥 DOC_CREATOR REQUEST RECEIVED")
         logger.info("=" * 80)
-        logger.info(f"👤 Client ID: {request.clientId}")
+        logger.info(f"👤 User ID: {request.userId}")
         logger.info(f"📋 Instruction: {request.instruction}")
         logger.info(f"📏 Document size: {len(request.documentContent)} bytes")
         
@@ -564,7 +564,7 @@ async def doc_creator_endpoint(request: DocCreatorRequest):
         - contentFormat: Always "html"
         - documentSummaries: Optional context from analyzed documents
         - conversationHistory: Optional previous conversation messages
-        - clientId: Unique client identifier
+        - userId: Unique user identifier (UUID)
     
     Returns:
         DocCreatorResponse with assistant's response and modified document

agent_states/agent_state.py

@@ -13,7 +13,7 @@ class AgentState(TypedDict):
     """
     # User interaction
     user_query: str
-    client_id: Optional[str]
+    user_id: Optional[str]
     conversation_history: List[Dict[str, str]]
     intermediate_steps: List[Dict[str, Any]]
     system_prompt: Optional[str]

agent_states/lawyer_messenger_state.py

@@ -12,7 +12,7 @@ class LawyerMessengerState(TypedDict):
     
     # Input
     conversation_history: List[dict]
-    client_id: str
+    user_id: str
     
     # Intermediate
     lawyers: Optional[List[dict]]  # Fetched from frontend API

langgraph_agent.py

@@ -113,10 +113,16 @@ class CyberLegalAgent:
                         args["jurisdiction"] = state.get("jurisdiction")
                         logger.info(f"🌍 Injecting jurisdiction: {args['jurisdiction']}")
                     
-                    # Inject client_id for message_lawyer tool
+                    # Inject user_id for message_lawyer tool
                     if tool_call['name'] == "message_lawyer":
-                        args["client_id"] = state.get("client_id")
-                        logger.info(f"👤 Injecting client_id: {args['client_id']}")
+                        args["user_id"] = state.get("user_id")
+                        logger.info(f"👤 Injecting user_id: {args['user_id']}")
+                    
+                    # Inject user_id for retrieve_lawyer_document tool
+                    if tool_call['name'] == "retrieve_lawyer_document":
+                        args["user_id"] = state.get("user_id")
+                        logger.info(f"📄 Injecting user_id for retrieve_lawyer_document: {args['user_id']}")
+                    
                     tool_call['name']="_" + tool_call['name']
                 
                 result = await tool_func.ainvoke(args)
@@ -126,10 +132,10 @@ class CyberLegalAgent:
         state["intermediate_steps"] = intermediate_steps
         return state
     
-    async def process_query(self, user_query: str, client_id: Optional[str] = None, jurisdiction: str = "Romania", conversation_history: Optional[List[Dict[str, str]]] = None, system_prompt: Optional[str] = None) -> Dict[str, Any]:
+    async def process_query(self, user_query: str, user_id: Optional[str] = None, jurisdiction: str = "Romania", conversation_history: Optional[List[Dict[str, str]]] = None, system_prompt: Optional[str] = None) -> Dict[str, Any]:
         initial_state = {
             "user_query": user_query,
-            "client_id": client_id,
+            "user_id": user_id,
             "conversation_history": conversation_history or [],
             "intermediate_steps": [],
             "relevant_documents": [],

prompts/main.py

@@ -71,6 +71,7 @@ Lawyer Jurisdiction: {jurisdiction}
 ### Available Tools
 1. **query_knowledge_graph**: Search legal documents to answer questions about law, regulations and directives.
 2. **search_web**: Search the web for current information, court decisions, or news that may not be in the knowledge graph.
+3. **retrieve_lawyer_document**: Retrieve the full content of a specific document from your document database. Use this tool when you need to inspect the detailed text, clauses, or specific provisions of a document mentioned in your documents_tree. But avoid it if the summary key details and actors are sufficient as it can introduce long document in the chat.
 
 ### Tool-Calling Process
 You operate in an iterative loop:
@@ -84,6 +85,7 @@ You operate in an iterative loop:
 2. Reference specific articles, paragraphs, recitals, and provisions from regulations and directives. Include precise legal citations.
 3. Analyze legal implications, potential interpretations, and relevant jurisprudence where applicable.
 4. Create a section at the end of your response called "References" that lists the source documents used to answer the user's question with full legal citations.
+5. If the user asks a specific detail about a document from your documents_tree, use the retrieve_lawyer_document tool, but avoid it if the summary key details and actors are sufficient as it can introduce long document in the chat.
 
 ### Tone
 - Professional and authoritative
@@ -94,13 +96,3 @@ You operate in an iterative loop:
 
 # Default system prompt (client-friendly for backward compatibility)
 SYSTEM_PROMPT = SYSTEM_PROMPT_CLIENT
-
-RESPONSE_FORMATTING_PROMPT = """### Task
-We aim at answering the user's question based on your common knowledge as well as the knowledge graph context.
-
-### Inputs
-**Knowledge Graph Context:**{context}
-**User Query:** {user_query}
-
-### Response format
-Answer in clear language and provide a direct answer to the user's question."""

structured_outputs/api_models.py

@@ -52,7 +52,7 @@ class DocumentsTree(BaseModel):
 
 class ChatRequest(BaseModel):
     """Chat request model"""
-    clientId: str = Field(..., description="Unique client identifier")
+    userId: str = Field(..., description="Unique user identifier (UUID)")
     message: str = Field(..., description="User's question")
     conversationHistory: Optional[List[Message]] = Field(default=[], description="Previous conversation messages")
     userType: Optional[str] = Field(default="client", description="User type: 'client' for general users or 'lawyer' for legal professionals")
@@ -102,7 +102,7 @@ class DocCreatorRequest(BaseModel):
     contentFormat: str = Field(default="html", description="Format of document content (always 'html')")
     documents_tree: Optional[DocumentsTree] = Field(default=None, description="Hierarchical tree of documents for context")
     conversationHistory: Optional[List[Message]] = Field(default=[], description="Previous conversation messages")
-    clientId: str = Field(..., description="Unique client identifier")
+    userId: str = Field(..., description="Unique user identifier (UUID)")
 
 
 class DocCreatorResponse(BaseModel):

subagents/lawyer_messenger.py

@@ -171,7 +171,7 @@ class LawyerMessengerAgent:
         API_KEY = os.getenv("CYBERLGL_API_KEY")
         
         payload = {
-            "clientId": state["client_id"],
+            "userId": state["user_id"],
             "lawyerId": lawyer_selection["lawyer_id"],
             "subject": lawyer_selection["subject"],
             "message": lawyer_selection["message"]
@@ -221,11 +221,11 @@ class LawyerMessengerAgent:
             state["message_sent"] = False
             return state
     
-    async def send_lawyer_message(self, conversation_history: List[dict], client_id: str) -> str:
+    async def send_lawyer_message(self, conversation_history: List[dict], user_id: str) -> str:
         """Main entry point: identify lawyer and send message"""
         result = await self.workflow.ainvoke({
             "conversation_history": conversation_history,
-            "client_id": client_id,
+            "user_id": user_id,
             "lawyers": None,
             "lawyer_selection": None,
             "message_sent": False,

utils/tools.py

@@ -219,14 +219,14 @@ async def message_lawyer() -> str:
         if lawyer_messenger_agent is None:
             raise ValueError("LawyerMessengerAgent not initialized. Please initialize it in agent_api.py")
         
-        # conversation_history and client_id will be injected by the agent from state
-        raise ValueError("conversation_history and client_id not provided - these should be injected by the agent")
+        # conversation_history and user_id will be injected by the agent from state
+        raise ValueError("conversation_history and user_id not provided - these should be injected by the agent")
     
     except Exception as e:
         return f"Error sending message to lawyer: {str(e)}"
 
 @tool
-async def _message_lawyer(conversation_history, client_id) -> str:
+async def _message_lawyer(conversation_history, user_id) -> str:
     """
     Send a message to a lawyer identified from the conversation.
     
@@ -245,7 +245,7 @@ async def _message_lawyer(conversation_history, client_id) -> str:
             raise ValueError("LawyerMessengerAgent not initialized. Please initialize it in agent_api.py")
         
         
-        return await lawyer_messenger_agent.send_lawyer_message(conversation_history, client_id)
+        return await lawyer_messenger_agent.send_lawyer_message(conversation_history, user_id)
     except Exception as e:
         return f"Error sending message to lawyer: {str(e)}"
 
@@ -339,9 +339,118 @@ async def _edit_document(doc_text: str, instruction: str, doc_summaries: List[st
         return f"Error editing document: {str(e)}"
 
 
+@tool
+async def retrieve_lawyer_document(file_path: str) -> str:
+    """
+    Retrieve a specific document from the lawyer's document database.
+    
+    This tool fetches a document by its file path from the lawyer's personal
+    document storage. It returns the extracted text, summary, actors, and key
+    details for the document.
+    
+    Use this tool when you need to inspect the full content of a document
+    mentioned in the documents_tree, such as when:
+    - User asks about specific clauses or details in a contract
+    - You need to verify information in a legal document
+    - User requests analysis of a specific document's content
+    
+    Args:
+        file_path: Path to the document from the documents_tree 
+                   (e.g., "Contracts/bail-commercial.pdf" or "note-juridique.pdf")
+    
+    Returns:
+        Document content including extracted_text, summary, actors, and key_details
+    """
+    try:
+        raise ValueError("user_id not provided - this should be injected by the agent")
+    except Exception as e:
+        return f"Error retrieving document: {str(e)}"
+
+
+@tool
+async def _retrieve_lawyer_document(
+    user_id: str,
+    file_path: str
+) -> str:
+    """
+    Retrieve a specific document from the lawyer's document database.
+    
+    Args:
+        user_id: UUID of the lawyer (injected from userId)
+        file_path: Path to the document (e.g., "Contracts/bail-commercial.pdf")
+        format: "text" (default) returns JSON with extracted_text, 
+                "raw" returns raw file bytes
+    
+    Returns:
+        Document content including extracted_text, summary, actors, and key_details
+    """
+    try:
+        import httpx
+        
+        # Check configuration from environment
+        lawyer_db_url = os.getenv("LAWYER_DB_URL")
+        cyberlgl_api_key = os.getenv("CYBERLGL_API_KEY")
+        
+        if not lawyer_db_url:
+            return "Error: LAWYER_DB_URL not configured in environment"
+        
+        if not cyberlgl_api_key:
+            return "Error: CYBERLGL_API_KEY not configured in environment"
+        
+        # Build URL - remove any trailing slash
+        base_url = lawyer_db_url.rstrip('/')
+        url = f"{base_url}/retrieve-lawyer-document"
+        
+        # Make request
+        async with httpx.AsyncClient() as client:
+            response = await client.get(
+                url,
+                params={
+                    "lawyer_id": user_id,
+                    "file_path": file_path,
+                    "format": 'text'
+                },
+                headers={
+                    "x-api-key": cyberlgl_api_key
+                },
+                timeout=30.0
+            )
+            
+            # Handle different status codes
+            if response.status_code == 404:
+                return f"Document not found: {file_path}"
+            elif response.status_code == 403:
+                return "Access denied: You do not have permission to access this document"
+            elif response.status_code != 200:
+                return f"Error retrieving document: HTTP {response.status_code}"
+            
+            # Parse JSON response
+            data = response.json()
+            
+            # Format the response for the LLM
+            output = [
+                f"📄 Document: {data.get('file_name', file_path)}",
+                "-" * 30,
+                "\nExtracted Text:",
+                data.get('extracted_text', 'No text available'),
+                "-" * 80
+            ]
+            
+            return "\n".join(output)
+            
+    except httpx.TimeoutError:
+        return "Error: Timeout while retrieving document"
+    except httpx.RequestError as e:
+        return f"Error: Failed to connect to document server: {str(e)}"
+    except json.JSONDecodeError:
+        return "Error: Invalid response from document server"
+    except Exception as e:
+        return f"Error retrieving document: {str(e)}"
+
+
 # Export tool sets for different user types
 tools_for_client_facade = [query_knowledge_graph, find_lawyers, message_lawyer, search_web, edit_document]
 tools_for_client = [_query_knowledge_graph, _find_lawyers, _message_lawyer, search_web, _edit_document]
-tools_for_lawyer_facade = [query_knowledge_graph, search_web, edit_document]
-tools_for_lawyer = [_query_knowledge_graph, search_web, _edit_document]
-tools = tools_for_client
+tools_for_lawyer_facade = [query_knowledge_graph, search_web, edit_document, retrieve_lawyer_document]
+tools_for_lawyer = [_query_knowledge_graph, search_web, _edit_document, _retrieve_lawyer_document]
+tools = tools_for_client