pa_agent

Personal Assistant Agent with LangGraph

A powerful conversational agent with retrieval-augmented generation, MCP server, long- and short-term memory, and tool integrations—powered by LangGraph, LangChain, Redis, PostgreSQL, and Pinecone.

---

Features

• Conversational Chat
  • Interactive CLI (app/run.py)
  • Programmatic access via Python SDK (langgraph_sdk)
• Retrieval-Augmented Generation
  • Ingest & query PDF, Markdown, HTML, CSV & DOCX into Pinecone
  • index_docs(name, path_or_url) & query_index(name, question, k) tools
• MCP Server
  • 25+ CoinMarketCap endpoints exposed
• Long-Term Memory
  • Profile, projects & instructions stored in PostgreSQL, namespaced by user
• Short-Term Memory
  • Rolling summary of recent conversation (pruned after 10+ turns)
• Tool Integrations
  • Web & Knowledge: web_fetch, wiki_search, tavily_search
  • File Handling: inspect_file, summarise_file, extract_tables, ocr_image, save_uploaded_file
  • Finance: get_stock_quote, get_stock_news
• Robustness
  • Automatic retries on transient OpenAI errors
  • Healthchecks on Redis & Postgres in Docker Compose

Project Structure

personal_assistant/
├── app/
│   ├── config.py
│   ├── run.py           # CLI entrypoint
│   ├── graph/
│   │   ├── assistant.py # StateGraph definition
│   │   ├── state.py     # ChatState schema
│   │   └── memory/      # summarization, schemas
│   ├── tools/           # external tool fns
│   ├── rag/             # RAG loaders & indexers
│   ├── mcp/             # MCP server logic
│   └── schemas/
├── requirements.txt
├── docker-compose.yml
├── langgraph.json
└── README.md

Prerequisites

• Python 3.11+
• Docker & Docker Compose (for containerized deployment)
• Redis & PostgreSQL (local or via Docker)
• OpenAI, Pinecone, Tavily & CoinMarketCap API keys

Getting Started

Local Setup

1. Clone the repo:

   git clone https://github.com/amanzoni1/pa_agent && cd personal_assistant
   

2. Create and activate a virtual environment:

   python -m venv .venv
   source .venv/bin/activate
   

3. Install dependencies:

   pip install -r requirements.txt
   

4. Create a .env file in project root:

   OPENAI_API_KEY=your_openai_key
   LANGSMITH_API_KEY=your_langsmith_key
   REDIS_URI=redis://localhost:6379
   POSTGRES_URI=postgresql://postgres:postgres@localhost:5432/personal_assistant?sslmode=disable
   PINECONE_API_KEY=your_pinecone_key
   TAVILY_API_KEY=your_tavily_key
   COINMARKETCAP_API_KEY=your_cmc_key
   

5. Ensure Redis and Postgres are running locally.

Command-Line Interface (CLI)

Interact via chat in terminal:

python -m app.run --thread-id <optional-uuid> --user-id <optional-your_id>

Commands:

• /memory: Show long-term memory (profile, projects, instructions) stored.
• /mcp: get all the tools available from the MCP server.
• /exit or Ctrl-D: Quit.

Docker Compose Deployment

1. Build your LangGraph image:

   langgraph build -t my-assistant
   

2. Launch via Docker Compose:

   docker compose up -d
   

3. Access:

   • API: http://localhost:8123
   • Swagger Docs: http://localhost:8123/docs

Python SDK Usage

Install the SDK:

pip install langgraph_sdk

Example:

import asyncio
from langgraph_sdk import get_client
from langchain_core.messages import HumanMessage

async def main():
    client = get_client(url="http://localhost:8123")
    thread = await client.threads.create()
    run = await client.runs.create(
        thread["thread_id"],
        "my-assistant",
        input={"messages": [HumanMessage(content="Hello!")]},
        config={"configurable": {"user_id": "you", "thread_id": thread["thread_id"]}},
    )
    final = await client.runs.join(thread["thread_id"], run["run_id"])
    msgs = final.get("messages", [])
    ai = msgs[1]["content"] if len(msgs) > 1 else None
    print("AI:", ai)

asyncio.run(main())