Agents

If you’re a coding agent (Cursor Agent, Codex, Claude Code, and friends), start with the repo-root AGENTS.md. That’s the canonical bootstrap.

Quick version

This service is an OpenAI-compatible proxy in front of Codex/ChatGPT, Gemini/Antigravity, and Claude Code — still no sk- keys.

Run it:

codex login && go run ./cmd/open-agent-api --host 127.0.0.1 --port 8088
# or: docker compose up --build -d

Check it’s alive:

curl -s http://127.0.0.1:8088/health
curl -s http://127.0.0.1:8088/v1/models | jq '.data[].id'

Cursor BYOK: use a dummy API key, a public https://<tunnel>/v1 base URL (not localhost), and a model from the catalog (e.g. gpt-5.6-terra). Ngrok overlay: docker compose -f docker-compose.yml -f docker-compose.ngrok.yml up -d. Tool wire details: tool conventions.

Please don’t:

• Invent model slugs — take them from internal/openai/models.go or live /v1/models
• Commit secrets, or casually bind-mount ~/.claude.json
• Change aliases in models.go without updating the model catalog

Useful paths: cmd/open-agent-api/, internal/openai/models.go, docker-compose*.yml, scripts/sync-antigravity-auth.sh.

The full do/don’t list and doc map live in AGENTS.md.