Module chat_inference 

End-to-end Webizen-gated chat inference with retrieval, streaming, and provenance.

Structs

ChatInferenceOptions

ChatInferenceResult

Functions

clear_cancel_inference

is_inference_cancelled

request_cancel_inference

run_chat_inference_for_agent

Run a local turn using a named roster agent. A pinned model is activated on demand before inference; the lifecycle implementation owns any resident mapping replacement. This is a cold control-path operation and never runs inside the decode/evaluator hot path.

run_chat_inference_full

run_chat_inference_with_options

stream_event_done

stream_event_error

stream_event_token

NDJSON stream events for Flutter FRB.

validate_axiom_preflight

Pre-flight axiom bounds check before KV prefill / orchestrator dispatch.