Autonomous GitHub issue resolution. Point AutoPR at a repo, pick an issue, and it indexes the codebase, drafts a fix, runs tests in a sandbox, and opens a pull request.
- Python 3.11+
- Node.js 18+
- Docker (for the sandbox agent)
- A Nia account (codebase indexing)
cp .env.example .envFill in .env:
| Variable | Description |
|---|---|
LLM_PROVIDER |
anthropic or openai |
ANTHROPIC_API_KEY |
Required if using Anthropic |
ANTHROPIC_MODEL |
Defaults to claude-sonnet-4-5 |
OPENAI_API_KEY |
Required if using OpenAI |
OPENAI_MODEL |
Defaults to gpt-5 |
NIA_API_KEY |
Nia API key for codebase indexing |
GITHUB_TOKEN |
Personal access token with repo scope |
AGENT_PR_COMMENTS_ENABLED |
Set true to post agent commentary as PR comments |
cd backend
pip install -r requirements.txt
uvicorn main:app --reloadRuns on http://localhost:8000.
cd frontend
npm install
npm run devRuns on http://localhost:5173.
- Open
http://localhost:5173 - Enter a GitHub repo URL (e.g.
https://github.com/owner/repo) - AutoPR indexes the codebase via Nia and loads open issues
- Select an issue, add optional context, and hit Start Resolution
- Watch the pipeline run live — indexing → code changes → tests → PR
Runs persist across page reloads. Cancel or dismiss any run from the dashboard.