bus-api-provider-inference

Inference API Provider

bus-api-provider-inference is the running-Bus HTTP surface for inference runtime capabilities. It keeps API controllers provider-neutral and delegates provider-specific runtime work through bus-integration-inference.

Run this provider on loopback or a protected internal network behind bus-api. The current provider handler is an internal component and does not perform public-client authentication by itself; production deployments must protect /api/internal/inference/* with Bus API routing, internal networking, and operator authorization. Configure the matching inference integration providers before using the capability endpoint in a deployment. For Ollama-backed deployments this means installing bus-integration-ollama from the same BusDK release set and setting BUS_INFERENCE_PROVIDER=ollama in the operator or service environment.

Available endpoints return 200 with JSON on success:

  • GET /healthz returns {"ok":true,"service":"bus-api-provider-inference"}.
  • GET /api/internal/inference/capabilities returns provider-neutral inference capability metadata with providers, events, and direct-bootstrap support.
BUS_INFERENCE_PROVIDER=ollama bus-api-provider-inference --addr 127.0.0.1:8094

In a second shell, verify the provider:

curl -fsS http://127.0.0.1:8094/healthz
curl -fsS http://127.0.0.1:8094/api/internal/inference/capabilities

The first check succeeds with ok: true; the second includes inference event names such as bus.inference.install.request.

For protected access through bus-api, enable the built-in provider by name and enable the matching module mount. Store the Bus API capability token in an untracked operator secret file before starting the service, then validate it is non-empty:

cd /path/to/deployment-repository
install -m 700 -d ./local
git check-ignore -q ./local/bus-api-capability-token || printf '%s\n' '/local/' >> .git/info/exclude
git check-ignore -q ./local/bus-api-capability-token
test -s ./local/bus-api-capability-token || openssl rand -hex 32 > ./local/bus-api-capability-token
chmod 600 ./local/bus-api-capability-token
test -s ./local/bus-api-capability-token
BUS_API_CAPABILITY_TOKEN="$(tr -d '\r\n' < ./local/bus-api-capability-token)"
test -n "$BUS_API_CAPABILITY_TOKEN"
bus-api serve --token "$BUS_API_CAPABILITY_TOKEN" --port 8080 \
  --provider inference \
  --enable-module inference

Keep bus-api serve running in that shell. In a second shell, set $BUS_API_BASE_URL to the local capability URL printed by bus-api, then verify the protected route:

cd /path/to/deployment-repository
BUS_API_CAPABILITY_TOKEN="$(tr -d '\r\n' < ./local/bus-api-capability-token)"
test -n "$BUS_API_CAPABILITY_TOKEN"
export BUS_API_BASE_URL="http://127.0.0.1:8080/${BUS_API_CAPABILITY_TOKEN}/v1"
curl -fsS "$BUS_API_BASE_URL/api/internal/inference/capabilities"

Use this provider on internal Bus routes for operator and deployment automation. Ollama-specific work belongs to bus-integration-ollama.

Using From .bus Files

Repository-local automation should start the provider on loopback with an explicit provider environment and let bus-api or the supervisor own public routing:

# same as: BUS_INFERENCE_PROVIDER=ollama bus-api-provider-inference --addr 127.0.0.1:8094
run command -- sh -c 'BUS_INFERENCE_PROVIDER=ollama exec bus-api-provider-inference --addr 127.0.0.1:8094'

Sources