MemoryPilot

⚡ Brief

A memory-powered AI agent with a live chatroom for conversational retrieval, using pgvector for vectors database, FastAPI backend and Vue frontend.

⚒️ Setup

remote llm

1. install dependencies

   cd backend
   uv add langchain-google-genai
   uv export --format requirements.txt > requirements.txt --no-hashes --no-annotate
   

2. environment variables (.env)

   # backend
   BACKEND_PORT="8000"
   BACKEND_BASE_URL="http://localhost:${BACKEND_PORT}"
   
   # database
   POSTGRES_USER="postgres"
   POSTGRES_PASSWORD="postgres123"
   POSTGRES_HOST="db"
   POSTGRES_PORT="5432"
   POSTGRES_DB="langchain"
   VECTOR_DB_URL="postgresql+asyncpg://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT}/${POSTGRES_DB}"
   VECTOR_DB_PROVIDER="postgres"
   VECTOR_DB_COLLECTION="documents"
   
   # model
   EMBEDDING_PROVIDER="google"
   EMBEDDING_MODEL="text-embedding-004"
   LLM_PROVIDER="google"
   LLM_MODEL="gemini-2.5-flash"
   GOOGLE_API_KEY="..."

3. build & run app

   docker compose --profile remote-llm up
   

local llm

1. install dependencies

   cd backend
   uv add langchain-ollama
   uv export --format requirements.txt > requirements.txt --no-hashes --no-annotate
   

2. environment variables (.env)

   # backend
   BACKEND_PORT="8000"
   BACKEND_BASE_URL="http://localhost:${BACKEND_PORT}"
   
   # database
   POSTGRES_USER="postgres"
   POSTGRES_PASSWORD="postgres123"
   POSTGRES_HOST="db"
   POSTGRES_PORT="5432"
   POSTGRES_DB="langchain"
   VECTOR_DB_URL="postgresql+asyncpg://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT}/${POSTGRES_DB}"
   VECTOR_DB_PROVIDER="postgres"
   VECTOR_DB_COLLECTION="documents"
   
   # model
   EMBEDDING_PROVIDER="ollama"
   EMBEDDING_MODEL="nomic-embed-text"
   LLM_PROVIDER="ollama"
   LLM_MODEL="llama3.1"
   OLLAMA_PORT="11434"
   OLLAMA_BASE_URL="http://ollama:${OLLAMA_PORT}"

3. build & run app

   docker compose --profile local-llm up
   

4. install local llm

   docker exec -it memorypilot-ollama-1 ollama pull nomic-embed-text
   docker exec -it memorypilot-ollama-1 ollama pull llama3.1
   

local test

1. install dependencies

   cd backend
   uv sync --group test
   uv add langchain-ollama
   uv export --format requirements.txt > requirements.txt --no-hashes --no-annotate
   

2. environment variables (.env)

   # backend
   BACKEND_PORT="8000"
   BACKEND_BASE_URL="http://localhost:${BACKEND_PORT}"
   
   # database
   POSTGRES_USER="postgres"
   POSTGRES_PASSWORD="postgres123"
   POSTGRES_HOST="db"
   POSTGRES_PORT="5432"
   POSTGRES_DB="langchain"
   VECTOR_DB_URL="postgresql+asyncpg://${POSTGRES_USER}:${POSTGRES_PASSWORD}@${POSTGRES_HOST}:${POSTGRES_PORT}/${POSTGRES_DB}"
   VECTOR_DB_PROVIDER="postgres"
   VECTOR_DB_COLLECTION="documents"
   
   # model
   EMBEDDING_PROVIDER="ollama"
   EMBEDDING_MODEL="nomic-embed-text"
   LLM_PROVIDER="ollama"
   LLM_MODEL="llama3.1"
   OPENAI_API_KEY="..."  # judge model is GPT-4o
   OLLAMA_PORT="11434"
   OLLAMA_BASE_URL="http://ollama:${OLLAMA_PORT}"

3. build & run app

   docker compose --profile local-test up
   

4. install local llm

   docker exec -it memorypilot-ollama-1 ollama pull nomic-embed-text
   docker exec -it memorypilot-ollama-1 ollama pull llama3.1