Sandboxed Repo Agent

A FastAPI app that answers questions about public GitHub repositories using the OpenAI Agents API public beta and Vercel Sandbox. It streams progress and returns an answer with RESULT.md, evidence.json, and events.jsonl downloads.

The deploy button requests both OpenAI keys. The public OpenAI Python SDK is installed from PyPI; no private SDK repository access is needed.

Local setup

You need Python 3.11 or later, Git, the Vercel CLI, a Vercel project with Sandbox access, and an OpenAI project with Agents API access.

git clone https://github.com/vercel-labs/openai-agents-api-fastapi.git
cd openai-agents-api-fastapi
python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements.txt -r requirements-dev.txt
cp .env.example .env

Set these values in .env:

Variable	Value
`OPENAI_API_KEY`	Application key with `api.agents.read`, `api.agents.write`, and `api.responses.write`
`OPENAI_EXECUTOR_API_KEY`	Separate environment key; set other permissions to None

Use keys from the same OpenAI organization, project, and user or service account. The agent is defined inline for each session; no AGENT_ID is needed.

Link your Vercel project and pull its local Sandbox credentials:

vercel login
vercel link
vercel env pull .env.local

This supplies VERCEL_OIDC_TOKEN for local Sandbox calls. Refresh it with the same pull command if it expires. The app loads .env.local before .env; remove blank or duplicate OpenAI key entries from .env.local if they mask the values in .env. Keep both files untracked. Deployed apps use the project's Vercel identity for Sandbox access.

Start FastAPI:

python -m uvicorn app:app --reload

Open localhost:8000, enter a public GitHub repository and a valid branch, tag, or commit, and ask a focused question. Check /api/health for configuration status; a completed inspection verifies the service connections.

For implementation details, see ARCHITECTURE.md. Coding agents should read AGENTS.md.