Repository navigation
What’s new
- Add a configurable direct OpenAI Responses API vision provider (
VISION_PROVIDER=openai). - Send camera JPEG frames and questions directly to the default
gpt-4omodel; preserve the Xiaozhi vision gateway as the default provider. - Add mocked API tests for image payloads, model configuration, response validation, missing credentials, HTTP failures, and secret redaction.
Docker
docker pull ghcr.io/zbbio/lab-robot:v0.2.0For GPT-4o, set VISION_PROVIDER=openai, OPENAI_API_KEY, and optionally OPENAI_MODEL in the local .env file, then recreate the Compose service.
The OpenAI provider was validated with mocked API responses; configure a valid OpenAI API key to make a live request. See the README for setup details.