Release 1.0.0
Initial capabilities include:
- Trajectory generation through native
POST /v1/infer, OpenAI-compatiblePOST /v1/chat/completions, and the binary gRPCPredictAPI. - Visual question answering through
POST /v1/vqa. - Automatic single-GPU profile selection with BF16, FP8, and W4A16 precision options on supported hardware.
- Sample-data and payload-building tools for runnable 16-image trajectory and single-image VQA requests.
- Per-trajectory rotations, kinematic comfort metrics, health checks, and Prometheus metrics.
For current operational limitations, refer to Known Issues and Limitations.