feat(chat): add LangChain memory service for conversation history

Summary

• Implement LangChain-based conversation memory with Summary + Buffer strategy
• Add /memory endpoints for n8n workflow integration
• Externalize all configuration to .env file
• Switch chatbot-rag workflow to use OpenAI API directly

Changes

New Features

1. LangChain Memory Service

• Strategy: Summary + Buffer
  • Buffer: Keep last 10 messages in memory
  • Summary: Auto-generate summary when > 6 new messages accumulate
  • Storage: Summary stored in conversations.summary column

2. Memory API Endpoints

Method	Endpoint	Description
GET	/chat/conversations/{id}/memory	Get conversation memory for LLM
POST	/chat/conversations/{id}/memory	Get memory with request body

3. Configuration Externalization

All constants moved to .env:

# LiteLLM
LITELLM_URL=https://litellm-proxy.mc-intern.com/v1/chat/completions
LITELLM_MODEL=gpt-5
LITELLM_API_KEY=sk-...

# Memory
MEMORY_BUFFER_SIZE=10
MEMORY_MAX_MESSAGE_LENGTH=4000
MEMORY_SUMMARY_THRESHOLD=6

# RAG
RAG_KVEC=5
RAG_KLEX=3

4. OpenAI Direct Integration

• Workflow chatbot-rag.json now calls OpenAI API directly
• No dependency on shared n8n credentials
• API key passed in Authorization header

Database Migration

New columns in conversations table:

Column	Type	Description
summary	TEXT	Summary of old messages
summary_message_count	INTEGER	Number of summarized messages

Run migration:

python Agent/scripts/migrate_conversation_summary.py

Files Changed

File	Change
Agent/scripts/migrate_conversation_summary.py	New - Migration script
Integration/config/constants.py	New - Externalized config
Integration/services/langchain_memory_service.py	New - Memory service
Integration/models/conversation.py	Modified - Add summary columns
Integration/schemas/chat.py	Modified - Add Memory schemas
Integration/chat_api.py	Modified - Add /memory endpoints
Integration/workflows/chatbot-rag.json	Modified - OpenAI direct
Agent/.env	Modified - Add new variables

Test Plan

• Run migration script
• Test /memory endpoint (GET)
• Test /memory endpoint (POST)
• Test conversation with 5+ messages
• Verify history is passed to LLM
• Verify AI responds with context awareness
• Verify database structure (user/agent alternation)
• Run Integration/tests/test_chat_memory.sh - All 8 tests pass

Architecture

User Message
     |
     v
POST /chat/conversations/{id}/messages
     |
     v
n8n Workflow (chatbot-rag)
     |
     +---> GET /chat/conversations/{id}/memory
     |          |
     |          v
     |     LangChainMemoryService
     |          |
     |          +---> Summary (if available)
     |          +---> Buffer (last 10 messages)
     |
     +---> POST /search/mixed (RAG)
     |
     v
Build Context: [Summary] + [Buffer] + [RAG Docs]
     |
     v
OpenAI API (gpt-5)
     |
     v
Response saved to DB