Repochat Action

Warning

This project is in early development and may contain bugs. Use with caution in production environments.

Chat with your repo in under 2 minutes using GitHub Actions on supported Cloud providers.

Features:

• ✨ A nice web UI
• Available as a GitHub Action
• 🔄 Sateless ingestion via API
• 💾 Optional state persisted to PostgreSQL or ChromaDB
• 🔐 Optional password on web UI

Requirements

• A supported cloud provider credentials ;
• An OpenAI API key.

Usage for GitHub Actions

Easily add RepoChat to your project using GitHub Actions:

name: "Deploy Repochat for this repo"

on:
  push:
    branches:
    - main

jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
    - uses: actions/checkout@v2
    - uses: flavienbwk/repochat-action@v0
      name: 'Deploy Repochat'
      id: deploy_repochat
      with:
        # All parameters not explicitly marked as "optional" are required
        dirs_to_scan: "."  # comma-separated glob dirs to analyze
        interface_password: ${{ secrets.INTERFACE_PASSWORD }}  # optional
        openai_api_key: ${{ secrets.OPENAI_API_KEY }}
        openai_model_type_inference: "gpt-4o-mini"
        openai_model_type_embedding : "text-embedding-3-small"
        provider_name: 'scaleway'
        provider_key_id: ${{ secrets.PROVIDER_KEY_ID }}
        provider_key_secret: ${{ secrets.PROVIDER_KEY_SECRET }}
        provider_project_id: ${{ secrets.PROVIDER_PROJECT_ID }}
        provider_default_region: 'fr-par'
        provider_default_zone: 'fr-par-2'

    - name: Get repochat domain
      run: echo "DOMAIN=${{ steps.deploy_repochat.outputs.domain }}" >> $GITHUB_OUTPUT
      id: repochat_domain

Name	Required	Secret	Description
dirs_to_scan	Yes	No	Comma-separated glob directories to analyze
openai_api_key	Yes	Yes	OpenAI API key for authentication
openai_model_type_inference	Yes	No	OpenAI model type for inference (e.g., "gpt-4o-mini")
openai_model_type_embedding	Yes	No	OpenAI model type for embedding (e.g., "text-embedding-3-small")
provider_name	Yes	No	Name of the cloud provider (e.g., 'scaleway')
provider_key_id	Yes	Yes	Cloud provider API key ID
provider_key_secret	Yes	Yes	Cloud provider API key secret
provider_project_id	Yes	Yes	Cloud provider project ID
provider_default_region	Yes	No	Default region for the cloud provider (e.g., 'fr-par')
provider_default_zone	Yes	No	Default zone for the cloud provider (e.g., 'fr-par-2')
interface_password	No	Yes	Optional password for the interface
pg_connection_string	No	Yes	Enables storage to external PG DB (format: 'postgresql://username:password@hostname:port/database_name')
cpu_limit	No	No	Default to 1000 (1 vCPU). Capabilities depend on the Cloud provider.
memory_limit	No	No	Default to 1024. Capabilities depend on the Cloud provider.
min_scale	No	No	Default to 1. Must be left to 1 if not using the PG connection.
max_scale	No	No	Default to 1. Must be left to 1 if not using the PG connection.

Get a practical implementation example with .github/workflows/push-deploy.yml.

You can restrict your OpenAI API key permissions to:

• Models: Read
• Model capabilities: Write

Supported Cloud providers

• Scaleway

  • Refer to Scaleway's documentation to generate API keys.

  • Additional required parameters:

    provider_name: 'scaleway'
    provider_key_id: ${{ secrets.PROVIDER_KEY_ID }}
    provider_key_secret: ${{ secrets.PROVIDER_KEY_SECRET }}
    provider_project_id: ${{ secrets.PROVIDER_PROJECT_ID }}
    provider_default_region: 'fr-par'  # example
    provider_default_zone: 'fr-par-2'  # example

  • Policies: ContainersFullAccess, ServerlessJobsFullAccess, FunctionsFullAccess

Other deployments

Deploy with Docker

👉 Deploy locally as a stateless API...

1. Copy and update env variables

   cp .env.example .env

2. Run the Docker container

   docker compose up --build -d

3. Inject data taking example on the Python or JS scripts

4. Access the app at http://localhost:3001

Development

👉 Run RepoChat for development...

1. Clone this repo

   [email protected]:flavienbwk/repochat-action.git

2. Copy and update env variables

   cp .env.example .env

3. Run the local stack

   make dev

4. Access the app at http://127.0.0.1:3000

Release Action

1. Increase repochat's version in ./package.json

2. Run build and commit latest edits:

   npm run build
   # git add && git commit && git push...

3. Merge on main

   This will create a release based on package.json and push the :latest Docker image.

Why not use Vercel ?

Vercel is very limited when it comes to deploying everything but JS. First, ChromaDB (and any sqlite-based library) is not supported in Vercel. Then, this project uses a FastAPI Python API that requires more storage than Vercel's 250MB bundle limit.