Principle:Langchain ai Langchain Streaming Events
| Knowledge Sources | |
|---|---|
| Domains | Streaming, Observability |
| Last Updated | 2026-02-11 00:00 GMT |
Overview
An event-driven streaming API that provides structured lifecycle events from all Runnables in a chain, enabling fine-grained observability of complex pipelines.
Description
While stream() provides chunks from a single model, astream_events() provides structured events from every Runnable in a composed chain. Events include start, stream (per-chunk), and end for each component. This enables monitoring retrieval, formatting, model generation, and parsing stages independently.
Events follow the format: on_{runnable_type}_{start|stream|end} with metadata including name, run_id, parent_ids, tags, and data.
Usage
Use astream_events() when building complex chains and you need visibility into intermediate steps, or when building UIs that display progress from multiple chain components.
Theoretical Basis
# Abstract event stream (not real code)
async for event in chain.astream_events(input, version="v2"):
if event["event"] == "on_chat_model_stream":
display(event["data"]["chunk"].content)
elif event["event"] == "on_retriever_end":
show_sources(event["data"]["output"])