Server Utilities, Demos & Docs

    Talk to a running server

    These call a running openeye serve at OPENEYE_SERVER_URL (default http://localhost:8000); each takes --server, -s <url>.

    CommandDescription
    openeye healthServer health and inference queue status
    openeye nebius-statsNebius VLM usage statistics (GET /nebius/stats)
    openeye server-config-getPrint the server's runtime config (GET /config)
    openeye server-config-set KEY VALUEUpdate one runtime config value (VALUE is parsed as JSON when possible; the rest of the config is kept)
    openeye health
openeye server-config-set vlm_model qwen/qwen3.5-9b

    openeye govern

    Visual governance: policies, presets, and the audit trail. Without --server these work on the local governance engine; with --server they act on a running server's /governance API.

    CommandDescription
    govern status [--server]Current governance status
    govern lsList available policies (built-in and plugins)
    govern enable NAME [--server]Enable a policy
    govern disable NAME [--server]Disable a policy
    govern presetsList presets
    govern load TARGET [--server]Load a preset or a custom YAML config
    govern validate FILEValidate a YAML governance config
    govern audit [--server] [--limit, -n 20]Recent audit trail
    govern violations [--server] [--limit, -n 20]Violations from the audit trail
    govern init [--domain, -d robotics] [--output, -o governance.yaml]Write a starter config (domains: robotics, desktop_agent, universal)

    See Governance API for the HTTP routes.

    openeye mlops

    Model lifecycle commands: registry and versions, promotion gates (promote, approve-promotion, reject-promotion), A/B tests and shadow mode, retraining pipelines, batch inference, feedback and annotations, export (ONNX, TensorRT, CoreML), lineage, evaluation, benchmarks, and validation tests. State is kept in YAML files under ~/.openeye/. Run openeye mlops --help or openeye docs for every subcommand and flag; MLOps API documents the matching HTTP routes on openeye serve.

    Demos

    CommandDescription
    openeye g1-demoUnitree G1 Safety Guardian demo
    openeye vlm-demoSend an image to a VLM via OpenRouter (needs OPENROUTER_API_KEY)

    g1-demo options: --transport (webcam, usb, rtsp, sdk, auto; default webcam), --host (G1 IP, default 192.168.123.161), --camera, -c (default 0), --control-mode (sdk, http, dry_run; default dry_run), --model, -m (default yolov8), --danger-m (0.5), --caution-m (1.5), --clear-duration (2.0 s), --max-fps (15), --video, -v, --demo.

    openeye g1-demo --control-mode dry_run --demo

    vlm-demo options: --image (default src/assets/demo/scene-warehouse.jpg, relative to a repo checkout), --model, -m (default qwen/qwen3-vl-235b:free), --prompt, -p, --all-models (compare all free VLM models side by side).

    openeye docs

    Prints the full CLI reference. Add --json, -j to get it wrapped in JSON ({"reference": "..."}) for AI agents.

    openeye docs --json | jq -r .reference
    ← openeye fleetOverview →