Skip to main content Module providers Copy item path Source anthropic openai CacheReadBySlot I1: per-slot attribution of Anthropic cache_read_input_tokens. Each field is None when that
slot did not carry a cache_control breakpoint on the request. Mirrors the Node SDK shape. ProviderToolSpec A tool specification in provider-facing format (parameters as a parsed JSON value). RuntimePolicy Per-model execution policy returned by providers.
Three-layer merge in RuntimeRunner: RuntimeOptions > provider > defaults. TokenUsage Token consumption for a single LLM call. StreamEvent Stream event emitted by providers. LLMProvider collect_message_from_stream ProviderRunState Opaque per-run state owned by the provider (e.g. OpenAI Responses continuation).