""" ChatManager Modul zur Verwaltung von AI-Chat-Workflows. Implementiert eine kompakte und modulare Architektur für die Verarbeitung von Benutzeranfragen, Agentenausführung und Ergebnisformatierung. """ import os import logging import json import re import uuid import base64 from datetime import datetime from typing import Dict, Any, List, Optional, Union # Notwendige Importe from connectors.connector_aichat_openai import ChatService from modules.chat_registry import get_agent_registry from modules.lucydom_interface import get_lucydom_interface from modules.chat_content_extraction import get_document_contents # Logger konfigurieren logger = logging.getLogger(__name__) class ChatManager: """ Verwaltet die Verarbeitung von Chat-Anfragen, Agentenausführung und die Integration von Ergebnissen in den Workflow. """ def __init__(self, mandate_id: int, user_id: int): """ Initialisiert den ChatManager mit Mandanten- und Benutzerkontext. Args: mandate_id: ID des aktuellen Mandanten user_id: ID des aktuellen Benutzers """ self.mandate_id = mandate_id self.user_id = user_id self.ai_service = ChatService() self.lucy_interface = get_lucydom_interface(mandate_id, user_id) self.agent_registry = get_agent_registry() self.agent_registry.set_ai_service(self.ai_service) ### Chat Management async def chat_run(self, user_input: Dict[str, Any], workflow_id: Optional[str] = None) -> Dict[str, Any]: """ Hauptfunktion zur Integration von Benutzeranfragen in den Workflow. Args: user_input: Dictionary mit Benutzeranfrage und Datei-IDs workflow_id: Optional - ID des Workflows (None für neue Workflows) Returns: Workflow-Objekt mit aktualisiertem Zustand """ # 1. Workflow initialisieren oder bestehenden laden workflow = self.workflow_init(workflow_id) # 2. User-Input in Message-Objekt transformieren und im Workflow speichern message_user = await self.chat_message_to_workflow("user", "", user_input, workflow) # 3. Projektleiter-Prompt erstellen und Antwort analysieren project_manager_response = await self.chat_prompt(message_user, workflow) obj_final_documents = project_manager_response.get("obj_final_documents", []) obj_workplan = project_manager_response.get("obj_workplan", []) obj_user_response = project_manager_response.get("obj_user_response", "") # 4. Speichere die Antwort als Message im Workflow und füge Log-Einträge hinzu response_message = { "role": "assistant", "agent_name": "project_manager", "content": obj_user_response } self.message_add(workflow, response_message) self.log_add(workflow, f"Geplante Ergebnisse: {self.parse_json2text(obj_final_documents)}") self.log_add(workflow, f"Arbeitsplan: {self.parse_json2text(obj_workplan)}") self.log_add(workflow, f"Info an den User: {obj_user_response}") # 5. Agenten gemäss Workplan ausführen obj_results = [] if obj_workplan: for task in obj_workplan: task_results = await self.agent_processing(task, workflow) obj_results.extend(task_results) # 6. Erstelle die finale Antwort mit den relevanten Dokumenten aus obj_final_documents final_message = await self.chat_final_message(obj_user_response, obj_final_documents, obj_results) self.message_add(workflow, final_message) # 7. Finalisiere den Workflow self.workflow_finish(workflow) return workflow async def chat_prompt(self, message_user: Dict[str, Any], workflow: Dict[str, Any]) -> Dict[str, Any]: """ Erstellt den Prompt für den Projektleiter und verarbeitet seine Antwort. Args: message_user: Message-Objekt mit Benutzeranfrage workflow: Aktuelles Workflow-Objekt Returns: Antwort des Projektleiters mit obj_final_documents, obj_workplan und obj_user_response """ # Verfügbare Agenten mit ihren Fähigkeiten abrufen available_agents = self.agent_profiles() # Erstelle eine Zusammenfassung des Workflows workflow_summary = await self.workflow_summarize(workflow, message_user) # Liste der aktuell verfügbaren Dokumente aus User-Input oder bereits generierten Dokumenten erstellen available_documents = self.available_documents_get(workflow, message_user) available_docs_str = json.dumps(available_documents, indent=2) # Erstelle den Prompt für den Projektleiter prompt = f""" Based on the user request and the provided documents, please analyze the requirements and create a processing plan. {message_user.get('content')} # Previous conversation history: {workflow_summary} # Available documents (currently in workflow): {available_docs_str} # Available agents and their capabilities: {self.parse_json2text(available_agents)} Please analyze the request and create: 1. A list of required result documents (obj_final_documents) 2. A plan for executing agents (obj_workplan) 3. A clear response to the user explaining what you're doing (obj_user_response) ## IMPORTANT RULES FOR THE WORKPLAN: 1. Each input document must either already exist (provided by the user or previously created by an agent) or be created by an agent before it's used. 2. If necessary, convert input documents to a suitable format using agents when the type doesn't match. 3. Do not define document inputs that don't exist or haven't been generated beforehand. 4. Create a logical sequence - earlier agents can create documents that are later used as inputs. 5. If the user has provided documents but hasn't clearly stated what they want, try to act according to the context. Your answer must be strictly in the JSON_OUTPUT format, with no additions before or after the JSON object. JSON_OUTPUT = {{ "obj_final_documents": [ FILEREF ], "obj_workplan": [ {{ "agent": "agent_name", # Name of an available agent "prompt": "Specific instructions to the agent, that he knows what to do with which documents and which output to provide." "output_documents": [ "label":"document label in the format 'filename.ext'", "prompt":"AI prompt to describe the content of the file" ], "input_documents": [ "label":"document label in the format 'filename.ext'", "file_id":id, # if refering to an existing document, provide file_id to select the correct file "content_part":"", # provide empty string, if all document contents to consider, otherwise the content_part of the document to focus on "prompt":"AI prompt to describe what data to extract from the file." ], # If no input documents are needed, include "input_documents" as an empty list }} # Multiple agent tasks can be added here and should build logically on each other ], "obj_user_response": "Information to the user about how his request will be solved." }} ## RULES for input_documents: 1. The user request refers to documents where "file_source" in available documents is "user". Those documents are in the focus for input 2. In case of redundant label in available documents, use document with highest sequence_nr if not specified differently ## STRICT RULES FOR document "label": 1. Every document label MUST include a proper file extension that matches the content type. 2. Use standard extensions like: - ".txt" for text files - ".md" for markdown files - ".csv" for comma-separated values - ".json" for JSON data - ".html" for HTML content - ".jpg" or ".png" for images - ".docx" for Word documents - ".xlsx" for Excel files - ".pdf" for PDF documents 3. Use descriptive filenames that indicate the document's purpose (e.g., "analysis_report.txt" rather than just "report.txt") 4. If you use label for an existing file """ # Rufe den AI-Service auf, um die Antwort des Projektleiters zu erhalten logger.debug(f"Planning prompt: {prompt}") project_manager_output = await self.ai_service.call_api([ { "role": "system", "content": "You are an experienced project manager who analyzes user requests and creates work plans. You pay very careful attention to ensure that all document dependencies are correct and that no non-existent documents are defined as inputs. The output follows strictly the specified format." }, { "role": "user", "content": prompt } ]) # Parsen der JSON-Antwort return self.parse_json_response(project_manager_output) async def chat_message_to_workflow(self, role: str, agent_name: str, chat_message: Dict[str, Any], workflow: Dict[str, Any]) -> Dict[str, Any]: """ Integriert Benutzereingaben in ein Message-Objekt inklusive Dateien mit vollständigen Inhalten. Args: chat_message: Eingabedaten "prompt"=str, "list_file_id"=[] Returns: Message-Objekt mit Inhalt und Dokumenten samt Inhalten """ logger.info(f"Message from {role} {agent_name} sent with {len(chat_message.get('list_file_id', []))} documents") logger.debug(f"message = {self.parse_json2text(chat_message)}.") # Nachrichteninhalt überprüfen message_content = chat_message.get("prompt", "") if isinstance(message_content, dict) and "content" in message_content: message_content = message_content["content"] # Wenn Nachrichteninhalt leer ist, kein Chat if role=="user" and (message_content is None or message_content.strip() == ""): logger.warning(f"Empty message, no chat") message_content = "(No user input received)" # Zusätzliche Dateien verarbeiten mit vollständigen Inhalten additional_fileids = chat_message.get("list_file_id", []) additional_files = await self.process_file_ids(additional_fileids) # Nachrichtenobjekt erstellen message_object = { "role": role, "agent_name": agent_name, "content": message_content, "documents": additional_files } message_object=self.message_add(workflow, message_object) logger.debug(f"message_user = {self.parse_json2text(message_object)}.") return message_object async def chat_final_message(self, obj_user_response: str, obj_final_documents: List[Dict[str, Any]], obj_results: List[Dict[str, Any]], ) -> Dict[str, Any]: """ Creates the final response message with review of proposed and delivered. Args: obj_user_response: Initial text response to the user obj_final_documents: List of expected response documents obj_results: List of generated result documents Returns: Complete message object with content and relevant documents """ # Find documents that match the obj_final_documents requirements matching_documents = [] for answer_spec in obj_final_documents: answer_label = answer_spec.get("label") # Find matching document in results for doc in obj_results: doc_name=self.get_filename(doc) # Check if this document matches the answer specification if doc_name == answer_label: content_ref = [] for c in doc.get("contents"): content_ref.append(c.get("summary")) doc_ref = { "label": doc_name, "content_summary": content_ref } matching_documents.append(doc_ref) break final_prompt = await self.ai_service.call_api([ {"role": "system", "content": "You are a project manager, who delivers results to a user."}, {"role": "user", "content": f""" Give the final short feedback to the user with reference to the initial statement (obj_user_response). Provide a list of delivered files (files_deliveded). If in the list of delivered files (files_delivered) some files from the original list (files_promised) are not available, then just give a comment on this, otherwise task is completed. Here the data: obj_user_response = {self.parse_json2text(obj_user_response)} files_promised = {self.parse_json2text(matching_documents)} files_deliveded = {self.parse_json2text(obj_user_response)} """ } ]) # Create basic message structure with proper fields logger.debug(f"FINAL PROMPT = {self.parse_json2text(final_prompt)}.") final_message = { "role": "assistant", "agent_name": "project_manager", "content": final_prompt, "documents": [] # DO NOT include the results documents, already with agents } logger.debug(f"FINAL MESSAGE = {self.parse_json2text(final_message)}.") return final_message ### Workflow def workflow_init(self, workflow_id: Optional[str] = None) -> Dict[str, Any]: """ Initializes a workflow or loads an existing one with round counting. Args: workflow_id: Optional - ID of the workflow to load Returns: Initialized workflow object """ current_time = datetime.now().isoformat() if workflow_id is None or not self.lucy_interface.get_workflow(workflow_id): # Create new workflow new_workflow_id = str(uuid.uuid4()) if workflow_id is None else workflow_id workflow = { "id": new_workflow_id, "mandate_id": self.mandate_id, "user_id": self.user_id, "name": f"Workflow {new_workflow_id[:8]}", "started_at": current_time, "messages": [], # Empty list - will be filled with references "message_ids": [], # Initialize empty message_ids list "logs": [], "data_stats": {}, "current_round": 1, "status": "running", "last_activity": current_time, "waiting_for_user": False } # Save to database - only the workflow metadata workflow_db = { "id": workflow["id"], "mandate_id": workflow["mandate_id"], "user_id": workflow["user_id"], "name": workflow["name"], "started_at": workflow["started_at"], "status": workflow["status"], "data_stats": workflow["data_stats"], "current_round": workflow["current_round"], "last_activity": workflow["last_activity"], "waiting_for_user": workflow["waiting_for_user"], "message_ids": workflow["message_ids"] # Include message_ids } self.lucy_interface.create_workflow(workflow_db) return workflow else: # Load existing workflow workflow = self.lucy_interface.load_workflow_state(workflow_id) # Ensure message_ids exists if "message_ids" not in workflow: # Initialize from existing messages workflow["message_ids"] = [msg["id"] for msg in workflow.get("messages", [])] # Update in database self.lucy_interface.update_workflow(workflow_id, {"message_ids": workflow["message_ids"]}) # Update status and increment round counter workflow["status"] = "running" workflow["last_activity"] = current_time workflow["waiting_for_user"] = False # Increment current_round if it exists, otherwise set it to 1 if "current_round" in workflow: workflow["current_round"] += 1 else: workflow["current_round"] = 1 # Update in database - only the relevant workflow fields workflow_update = { "status": workflow["status"], "last_activity": workflow["last_activity"], "waiting_for_user": workflow["waiting_for_user"], "current_round": workflow["current_round"] } self.lucy_interface.update_workflow(workflow_id, workflow_update) return workflow def workflow_finish(self, workflow: Dict[str, Any]) -> Dict[str, Any]: """ Finalizes a workflow and sets the status to 'completed'. Args: workflow: Workflow object Returns: Updated workflow object """ # Prepare workflow update data workflow_update = { "status": "completed", "last_activity": datetime.now().isoformat(), "waiting_for_user": True } # Update the workflow object in memory workflow["status"] = workflow_update["status"] workflow["last_activity"] = workflow_update["last_activity"] workflow["waiting_for_user"] = workflow_update["waiting_for_user"] # Save workflow state to database - only relevant fields, not the messages list self.lucy_interface.update_workflow(workflow["id"], workflow_update) return workflow async def workflow_summarize(self, workflow: Dict[str, Any], message_user: Dict[str, Any]) -> str: """ Erstellt eine Zusammenfassung des Workflows ohne die aktuelle User-Message. Args: workflow: Workflow-Objekt prompt: Anweisungen zur Erstellung der Zusammenfassung Returns: Zusammenfassung des Workflows """ if not workflow or "messages" not in workflow or not workflow["messages"]: return "" # die erste Message # Nachrichten in umgekehrter Reihenfolge durchgehen (neueste zuerst) messages = sorted(workflow["messages"], key=lambda m: m.get("sequence_no", 0), reverse=False) summary_parts = [] for message in messages: if True: # including user message, excluding would be: message["id"] != message_user["id"]: message_summary = await self.message_summarize(message) summary_parts.append(message_summary) return "\n\n".join(summary_parts) ### Agents def agent_profiles(self) -> List[Dict[str, Any]]: """ Ruft Informationen über alle verfügbaren Agenten ab. Returns: Liste mit Informationen über alle verfügbaren Agenten """ return self.agent_registry.get_agent_infos() async def agent_input_documents(self, doc_input_list: List[Dict[str, Any]], workflow: Dict[str, Any]) -> List[Dict[str, Any]]: """ Prepares input documents for an agent, sorted with newest first. Args: doc_input_list: List of required input documents as specified by the project manager workflow: Workflow object Returns: Prepared input documents for the agent, sorted with newest first """ prepared_inputs = [] # Sortiere die Workflow-Nachrichten nach Sequenznummer (absteigend) sorted_messages = sorted( workflow.get("messages", []), key=lambda m: m.get("sequence_no", 0), reverse=True ) for doc_spec in doc_input_list: doc_filename = doc_spec.get("label","") doc_file_id = doc_spec.get("file_id","") found_doc = None # Search for the document in sorted workflow messages (newest first) for message in sorted_messages: for doc in message.get("documents", []): if (doc_file_id!="" and doc_file_id==doc.get("file_id")) or (doc_filename!="" and self.get_filename(doc) == doc_filename): found_doc = doc break if found_doc: break if found_doc: # Process document for agent based on the specification processed_doc = await self.process_document_for_agent(found_doc, doc_spec) prepared_inputs.append(processed_doc) else: logger.warning(f"Document with label '{doc_filename}', file_id '{doc_file_id}' not found in workflow") return prepared_inputs async def process_document_for_agent(self, document: Dict[str, Any], doc_spec: Dict[str, Any]) -> Dict[str, Any]: """ Processes a document for an agent based on the document specification. Uses AI to extract relevant content from the document based on the specification. Args: document: The document to process doc_spec: The document specification from the project manager Returns: Processed document with AI-extracted content """ processed_doc = document.copy() part_spec = doc_spec.get("content_part","") # Process each content item in the document if "contents" in processed_doc: processed_contents = [] for content in processed_doc["contents"]: # Check if part required if part_spec != "" and part_spec != content.get("name"): continue # Get the data from the content data = content.get("data", "") processed_content = content.copy() # Check if content data is base64 encoded is_base64 = content.get("metadata", {}).get("base64_encoded", False) try: # Use the AI service to process the document content according to the prompt from the project manager for the document specification summary = doc_spec.get("prompt", "Extract the relevant information from this document") ai_prompt = f""" # Please process the following document content according to this instruction: {summary} # Document content: {data} # Extract and provide only the relevant information as requested. """ # Call the AI service to process the content processed_data = await self.ai_service.call_api([ {"role": "system", "content": "You are a document processing assistant. Extract only the relevant information as requested."}, {"role": "user", "content": ai_prompt} ]) # DO NOT change the original data field # processed_content["data"] unchanged processed_content["data_extracted"] = processed_data processed_content["metadata"]["ai_processed"] = True except Exception as e: logger.error(f"Error processing document content with AI: {str(e)}") # Fall back to original content if AI processing fails processed_content["data_extracted"] = "(no information)" processed_contents.append(processed_content) processed_doc["contents"] = processed_contents return processed_doc async def agent_processing(self, task: Dict[str, Any], workflow: Dict[str, Any]) -> List[Dict[str, Any]]: """ Process a single agent task from the workflow. Optimized for the task-based approach where all agents implement process_task. Args: task: The task definition containing agent name, prompt, and document specifications workflow: The current workflow object Returns: List of document objects created by the agent """ # Extract task information agent_name = task.get("agent") agent_prompt = task.get("prompt", "") # Log the current step output_labels = [d.get("label", "unknown") for d in task.get("output_documents", [])] step_info = f"Agent '{agent_name}' to create {', '.join(output_labels)}." self.log_add(workflow, step_info) # Check if prompt is empty if agent_prompt == "": logger.warning("Empty prompt, no task to do") return [] # Get agent from registry agent = self.agent_registry.get_agent(agent_name) if not agent: logger.error(f"Agent '{agent_name}' not found") return [] # Prepare output document specifications output_specs = [] for doc in task.get("output_documents", []): output_spec = { "label": doc.get("label"), "description": doc.get("prompt", "") } output_specs.append(output_spec) # Prepare input documents for the agent input_documents = await self.agent_input_documents(task.get('input_documents', []), workflow) # Create a standardized task object for the agent agent_task = { "task_id": str(uuid.uuid4()), "workflow_id": workflow.get("id"), "prompt": agent_prompt, "input_documents": input_documents, "output_specifications": output_specs, "context": { "workflow_round": workflow.get("current_round", 1), "agent_type": agent_name, "timestamp": datetime.now().isoformat() } } # Execute the agent with the standardized task try: # Process the task using the agent's standardized interface logger.debug("TASK: "+self.parse_json2text(agent_task)) logger.debug(f"Agent '{agent_name}' AI service available: {agent.ai_service is not None}") agent_results = await agent.process_task(agent_task) logger.debug(f"Agent '{agent_name}' completed task. RESULT: {self.parse_json2text(agent_results)}") # Log the agent response self.log_add( workflow, f"Agent '{agent_name}' completed task. Feedback: {agent_results.get('feedback', 'No feedback provided')}" ) # Store produced files and prepare input object for message agent_inputs = { "prompt": agent_results.get("feedback", ""), "list_file_id": self.agent_save_documents(agent_results) } # Create a message in the workflow with the agent's response agent_message = await self.chat_message_to_workflow("assistant", agent_name, agent_inputs, workflow) logger.debug(f"Agent result = {self.parse_json2text(agent_message)}.") return agent_message.get("documents", []) except Exception as e: error_msg = f"Error executing agent '{agent_name}': {str(e)}" logger.error(error_msg, exc_info=True) # Add exc_info=True to get full traceback self.log_add(workflow, error_msg, level="error") return [] def agent_save_documents(self, agent_results: Dict[str, Any]) -> List[int]: """ Saves all documents from agent results as files and returns a list of file IDs. Enhanced to handle the standardized document format from agents. Args: agent_results: Dictionary containing agent feedback and documents Returns: List of file IDs for the saved documents """ file_ids = [] # Extract documents from agent results documents = agent_results.get("documents", []) for doc in documents: try: # Extract label (filename) and content label = doc.get("label", "unnamed_file.txt") content = doc.get("content", "") # Split label into name and extension name, ext = os.path.splitext(label) if ext.startswith('.'): ext = ext[1:] # Remove leading dot elif not ext: # If no extension is provided, default to .txt for text content ext = "txt" label = f"{label}.{ext}" # Determine if content is base64 encoded is_base64 = False if isinstance(content, dict) and content.get("metadata", {}).get("base64_encoded", False): is_base64 = True content = content.get("data", "") # Convert content to bytes if isinstance(content, str): if is_base64: # Decode base64 to bytes try: file_content = base64.b64decode(content) except Exception as e: logger.warning(f"Failed to decode base64 content: {str(e)}") file_content = content.encode('utf-8') else: # Convert text to bytes file_content = content.encode('utf-8') else: # Already bytes file_content = content # Save file to database file_meta = self.lucy_interface.save_uploaded_file(file_content, label) if file_meta and "id" in file_meta: file_id = file_meta["id"] file_ids.append(file_id) logger.info(f"Saved document '{label}' with file ID: {file_id}") else: logger.warning(f"Failed to save document '{label}'") except Exception as e: logger.error(f"Error saving document from agent results: {str(e)}") # Continue with other documents instead of failing continue return file_ids ### Messages def message_add(self, workflow: Dict[str, Any], message: Dict[str, Any]) -> Dict[str, Any]: """ Adds a message to the workflow and updates last_activity. Saves the message in the database and updates the workflow with references. Args: workflow: Workflow object message: Message to be saved Returns: Added message """ current_time = datetime.now().isoformat() # Ensure messages list exists if "messages" not in workflow: workflow["messages"] = [] # Generate new message ID if not present if "id" not in message: message["id"] = f"msg_{str(uuid.uuid4())}" # Add workflow ID and timestamps message["workflow_id"] = workflow["id"] message["started_at"] = current_time message["finished_at"] = current_time # Set sequence number message["sequence_no"] = len(workflow["messages"]) + 1 # Ensure required fields are present if "role" not in message: # Set a default role based on agent_name message["role"] = "assistant" if message.get("agent_name") else "user" if "agent_name" not in message: message["agent_name"] = "" # Set status message["status"] = "completed" # Add message to workflow workflow["messages"].append(message) # Ensure message_ids list exists if "message_ids" not in workflow: workflow["message_ids"] = [] # Add message ID to the message_ids list workflow["message_ids"].append(message["id"]) # Update workflow status workflow["last_activity"] = current_time # Save to database - first the message itself self.lucy_interface.create_workflow_message(message) # Then save the workflow with updated references workflow_update = { "last_activity": current_time, "message_ids": workflow["message_ids"] # Update the message_ids field } self.lucy_interface.update_workflow(workflow["id"], workflow_update) return message async def message_summarize(self, message: Dict[str, Any]) -> str: """ Erstellt eine Zusammenfassung einer Nachricht einschließlich ihrer Dokumente. Args: message: Zu summarisierende Nachricht prompt: Anweisungen zur Erstellung der Zusammenfassung Returns: Zusammenfassung der Nachricht """ role = message.get("role", "undefined") agent_name = message.get("agent_name", "") content = message.get("content", "") try: content_summary = await self.ai_service.call_api([ {"role": "system", "content": f"You are a chat message summarizer. Create a very concise summary (2-3 sentences, maximum 300 characters)"}, {"role": "user", "content": content} ]) except Exception as e: logger.error(f"Fehler bei der Zusammenfassung: {str(e)}") content_summary = content[:200] + "..." # Dokumente zusammenfassen docs_summary = "" if "documents" in message and message["documents"]: docs_list = [] for i, doc in enumerate(message["documents"]): doc_name = self.get_filename(doc) docs_list.append(doc_name) if docs_list: docs_summary = "\nDocuments:" + "\n- ".join(docs_list) return f"[{role} {agent_name}]: {content_summary}{docs_summary}" async def message_summarize_content(self, content: Dict[str, Any]) -> str: """ Generates a summary for a content item using AI. Args: content: Content item to summarize (already processed by get_document_contents) Returns: Brief summary of the content """ # Extract relevant information data = content.get("data", "") content_type = content.get("content_type", "text/plain") is_text = content.get("metadata", {}).get("is_text", False) try: summary = await self.ai_service.call_api([ {"role": "system", "content": "You are a content summarizer. Create very concise summary (1-2 sentences, maximum 200 characters) about this file."}, {"role": "user", "content": f"Summarize this {content_type} content briefly:\n\n{data}"} ]) return summary except Exception as e: logger.error(f"Error generating content summary: {str(e)}") return f"Text content ({content_type})" ### Documents async def process_file_ids(self, file_ids: List[int]) -> List[Dict[str, Any]]: """ Processes a list of File-IDs and returns the corresponding file objects as a list of Document objects. Loads all contents directly and adds summaries to each content item. Args: file_ids: List of file IDs Returns: List of Document objects with contents and summaries """ documents = [] logger.info(f"Processing {len(file_ids)} files") for file_id in file_ids: try: # Check if the file exists file = self.lucy_interface.get_file(file_id) if not file: logger.warning(f"File with ID {file_id} not found") continue # Check if file belongs to the current mandate if file.get("mandate_id") != self.mandate_id: logger.warning(f"File {file_id} does not belong to mandate {self.mandate_id}") continue # Load file content file_content = self.lucy_interface.get_file_data(file_id) if file_content is None: logger.warning(f"No content found for file with ID {file_id}") continue # Create document file_name_ext = file.get("name") document = { "id": f"doc_{str(uuid.uuid4())}", "file_id": file_id, "name": os.path.splitext(file_name_ext)[0] if os.path.splitext(file_name_ext)[0] else "noname", "ext": os.path.splitext(file_name_ext)[1][1:] if os.path.splitext(file_name_ext)[1] else "bin", "data": base64.b64encode(file_content).decode('utf-8'), # Add file data as base64 "contents": [] } # Extract contents contents = get_document_contents(file, file_content) # Add summaries to each content item for content in contents: content["summary"] = await self.message_summarize_content(content) document["contents"] = contents logger.info(f"File {file.get('name', 'unnamed')} (ID: {file_id}) loaded with {len(contents)} contents and summaries") documents.append(document) except Exception as e: logger.error(f"Error processing file {file_id}: {str(e)}") # Continue with remaining files instead of failing continue return documents def available_documents_get(self, workflow: Dict[str, Any], message_user: Dict[str, Any]) -> List[Dict[str, Any]]: """ Determines all currently available documents from user input and already generated documents. Args: message_user: Current message from the user workflow: Current workflow object Returns: List with information about all available documents, sorted by message sequence_nr in descending order """ available_docs = [] if "messages" in workflow and workflow["messages"]: for message in workflow["messages"]: message_id = message.get("id", "unknown") sequence_nr = message.get("sequence_no", 0) # Determine source source = "user" if message_id == message_user.get("id") else "workflow" # Process documents in this message if "documents" in message and message["documents"]: for doc in message["documents"]: # Get filename using our helper method filename = self.get_filename(doc) file_id = doc.get("file_id") # Extract summaries from all contents content_summaries = [] for content in doc.get("contents", []): content_summaries.append({ "content_part": content.get("name","noname"), "metadata": content.get("metadata",""), "summary": content.get("summary","No summary"), }) # Create document info doc_info = { "sequence_nr": sequence_nr, "file_source": source, "file_id": file_id, "message_id": message_id, "label": filename, "content_summary_list": content_summaries, } available_docs.append(doc_info) # Sort by message sequence_nr in descending order (newest first) available_docs.sort(key=lambda x: x["sequence_nr"], reverse=True) logger.info(f"Available documents: {len(available_docs)}") return available_docs def save_document_to_file(self, document: Dict[str, Any]) -> Optional[int]: """ Speichert ein Document als Datei in der Datenbank und gibt die File-ID zurück. Args: document: Document-Objekt mit Inhalten Returns: File-ID oder None bei Fehler """ try: if not document or "contents" not in document or not document["contents"]: logger.warning("Dokument hat keine Inhalte zum Speichern") return None # Nimm den ersten Inhalt als Hauptinhalt main_content = document["contents"][0] name = main_content.get("name", "document") content_type = main_content.get("content_type", "text/plain") data = main_content.get("data", b"") # Binäre Daten sicherstellen if isinstance(data, str): data = data.encode('utf-8') # Datei in der Datenbank speichern file_meta = self.lucy_interface.save_uploaded_file(data, name) if file_meta and "id" in file_meta: # Aktualisiere das Document mit der File-ID document["file_id"] = file_meta["id"] return file_meta["id"] return None except Exception as e: logger.error(f"Fehler beim Speichern des Dokuments als Datei: {str(e)}") return None def add_document_to_message(self, message: Dict[str, Any], document: Dict[str, Any]) -> Dict[str, Any]: """ Fügt ein Document zu einer Nachricht hinzu. Args: message: Nachricht, zu der das Dokument hinzugefügt werden soll document: Hinzuzufügendes Document Returns: Aktualisierte Nachricht """ # Sicherstellen, dass die Dokumente-Liste existiert if "documents" not in message: message["documents"] = [] # Document hinzufügen message["documents"].append(document) return message ### Tools def get_filename(self, document: Dict[str, Any]) -> str: name = document.get("name", "unnamed") ext = document.get("ext", "") if ext: return f"{name}.{ext}" return name def log_add(self, workflow: Dict[str, Any], message: str, level: str = "info", agent_id: Optional[str] = None, agent_name: Optional[str] = None) -> str: """ Fügt einen Log-Eintrag zum Workflow hinzu und loggt diesen auch im Logger. Args: workflow: Workflow-Objekt message: Log-Nachricht level: Log-Level (info, warning, error) agent_id: Optional - ID des Agenten agent_name: Optional - Name des Agenten Returns: ID des erstellten Log-Eintrags """ # Sicherstellen, dass Logs-Liste existiert if "logs" not in workflow: workflow["logs"] = [] # Log-ID generieren log_id = f"log_{str(uuid.uuid4())}" # Log-Eintrag erstellen log_entry = { "id": log_id, "workflow_id": workflow["id"], "message": message, "type": level, "timestamp": datetime.now().isoformat(), "agent_id": agent_id, "agent_name": agent_name } # Log zum Workflow hinzufügen workflow["logs"].append(log_entry) # In Datenbank speichern self.lucy_interface.create_workflow_log(log_entry) # Auch im Logger loggen if level == "info": logger.info(f"Workflow {workflow['id']}: {message}") elif level == "warning": logger.warning(f"Workflow {workflow['id']}: {message}") elif level == "error": logger.error(f"Workflow {workflow['id']}: {message}") return log_id def parse_json2text(self, json_obj: Any) -> str: """ Konvertiert ein JSON-Objekt in eine lesbare Textdarstellung. Args: json_obj: Zu konvertierendes JSON-Objekt Returns: Formatierte Textdarstellung """ if not json_obj: return "Keine Daten vorhanden" try: # Formatieren mit Einrückung für bessere Lesbarkeit return json.dumps(json_obj, indent=2, ensure_ascii=False) except Exception as e: logger.error(f"Fehler bei JSON-Konvertierung: {str(e)}") return str(json_obj) def parse_json_response(self, response_text: str) -> Dict[str, Any]: """ Parst die JSON-Antwort aus einem Text. Args: response_text: Text mit JSON-Inhalt Returns: Geparste JSON-Daten """ try: # Extrahiere JSON aus dem Text (falls mit anderen Inhalten vermischt) json_start = response_text.find('{') json_end = response_text.rfind('}') + 1 if json_start >= 0 and json_end > json_start: json_str = response_text[json_start:json_end] return json.loads(json_str) else: # Versuche den gesamten Text zu parsen return json.loads(response_text) except json.JSONDecodeError as e: logger.error(f"JSON-Parse-Fehler: {str(e)}") # Fallback: Leere Struktur zurückgeben return { "obj_final_documents": [], "obj_workplan": [], "obj_user_response": "Sorry, I could not parse your data." } # Singleton-Factory für den ChatManager _chat_managers = {} def get_chat_manager(mandate_id: int = 0, user_id: int = 0) -> ChatManager: """ Gibt einen ChatManager für den angegebenen Kontext zurück. Wiederverwendet bestehende Instanzen. Args: mandate_id: ID des Mandanten user_id: ID des Benutzers Returns: ChatManager-Instanz """ context_key = f"{mandate_id}_{user_id}" if context_key not in _chat_managers: _chat_managers[context_key] = ChatManager(mandate_id, user_id) return _chat_managers[context_key]