LOPETools

Setup

To clone repo along with submodules, run

git clone --recursive [email protected]:lopentu/lopetools.git

Make sure to create .env in the root directory and add OPENAI_API_KEY to .env if you want to use OpenAI models:

OPENAI_API_KEY="<YOUR_KEY_HERE>"

To install backend requirements, run

pip install poetry==1.5.1
poetry install

To install frontend requirements, run

cd web
npm install

Quick start

To just look at usage examples, check out notebooks/langchain.ipynb

Structure and starting up the frontend and backend

The frontend is a React app in web/ and the backend is split into two parts in src/api/.

1. The frontend directly calls a FastAPI app that hosts a LangChain 🦜️🔗 agent in api.main:app
2. The SenseTagTool 🏷️ is hosted in a separate FastAPI app in api.tagger.main:app and is called by the LangChain agent.

Here's an unnecessarily detailed diagram of the flow:

sequenceDiagram
    box Frontend
    participant front as React App
    end
    box Backend
    participant agent as LangChain Agent
    participant tagger as SenseTagTool
    end
    front->>agent: POST /agent
    agent->>tagger: GET /tagger
    tagger->>agent: 200 OK
    agent->>front: 200 OK

Loading

I tried combining the agent and tagger into one app, but the agent hangs when it calls the tagger. I must be doing something wrong. 🤷‍♂️

Use something like tmux to start the frontend, agent, and tagger in separate terminals so they can run even when you close the terminal.

To start the frontend, run

cd web
npm run dev

To start the agent, run

cd src
uvicorn api.main:app --reload --port 8003 --host 0.0.0.0

To start the tagger, run

cd src
uvicorn api.tagger.main:app --reload --port 3001 

Docker

Because of the way the host server is setup, I can't get docker-compose to work. So I have to run the app without Docker. 😭