compose_chat_pipeline(...)

Compose chat ingestion primitives in the safe default order so action writes, summarization, and callback fan-out stay consistent.

Use compose_chat_pipeline(...) when you want one callback (pipeline.on_actions) that wires context storage and agent-context writing without callback clobbering.

Import

from autoplay_sdk import compose_chat_pipeline

Signature

compose_chat_pipeline(
    *,
    llm,
    threshold=10,
    lookback_seconds=300.0,
    max_actions=20,
    write_actions=None,
    overwrite_with_summary=None,
    debounce_ms=0,
)

• llm: async callable (prompt: str) -> str.
• threshold: summarizer trigger (action count).
• lookback_seconds, max_actions: AsyncContextStore bounds.
• write_actions: optional push callback for raw action text.
• overwrite_with_summary: optional callback invoked when summary is emitted.
• debounce_ms: forwarded to AsyncAgentContextWriter.

Minimal setup

from autoplay_sdk import compose_chat_pipeline

pipeline = compose_chat_pipeline(
    llm=my_async_llm,
    threshold=10,
    lookback_seconds=300,
    max_actions=20,
    write_actions=my_write_actions_cb,
    overwrite_with_summary=my_overwrite_cb,
)

client.on_actions(pipeline.on_actions)

Ordering guarantee

pipeline.on_actions(payload) runs in canonical order:

1. context_store.add(payload) updates retrieval state.
2. agent_writer.add(payload) handles push and summary logic.

Summary callbacks are fanned out so both context-store and writer-side overwrite paths receive updates. This prevents accidental callback overwrite behavior in custom wiring.

When to use this helper

• You want the default chat ingestion wiring without hand-rolling callback orchestration.
• You need predictable summary propagation to both storage and UI push callbacks.