LLMManagerMixin

Run on new output token.

Only available when streaming is enabled.

For both chat models and non-chat models (legacy text completion LLMs).

Run when LLM ends running.

Run when LLM errors.

Run on each protocol event from stream_events(version="v3").

Also fires for the async equivalent (astream_events(version="v3")).

Fires once per MessagesData event — message-start, per-block content-block-start / content-block-delta / content-block-finish, and message-finish. Analogous to on_llm_new_token in v1 streaming, but at event granularity rather than chunk: a single chunk can map to multiple events (e.g. a content-block-start plus its first content-block-delta), and lifecycle boundaries are explicit.

Fires uniformly whether the provider emits events natively via _stream_chat_model_events or goes through the chunk-to-event compat bridge. Observers see the same event stream regardless of how the underlying model produces output.

Not fired from v1 stream() / astream(); for those, keep using on_llm_new_token. Purely additive — on_chat_model_start, on_llm_end, and on_llm_error still fire around a v2 call as they do around a v1 call.