Skip to main content
This page describes the planned Thread type for aion-langgraph. The Thread object is the main fluent surface for replying, posting, looking up history, and emitting streaming intents without forcing authors to build raw A2A messages by hand.

​
Properties

PropertyPurpose
idCurrent messaging thread or context identifier
parent_idParent thread or context identifier when one exists
networkOriginating network or distribution identifier
default_reply_targetCanonical reply target derived from the inbound event

​
Methods

MethodPurpose
reply(content, *, metadata=None)Add a durable reply to the current thread
post(content, *, target=None, metadata=None)Create an explicit outbound post action
history(limit=20, before=None)Request recent conversation history through the control plane
typing()Emit a stream-only typing or progress intent
reply(...) and post(...) should accept the same content categories:
  • plain text
  • a provider-neutral card document
  • an async iterator of text chunks for streaming replies
  • an explicit low-level message builder when the author needs more control

​
reply(...)

reply(...) is intended for the most common case: answer in the same place the inbound message arrived. 
async def node(state: State, runtime: Runtime[AionContext]) -> dict:
    await runtime.context.thread.reply("Thanks. I'm looking into that now.")
    return {}
When content is an async iterator, the SDK should:
  1. emit stream deltas during SendStreamingMessage
  2. accumulate the durable reply in the same request-scoped response buffer
That keeps streamed output and final reply compilation aligned.

​
post(...)

post(...) is intended for explicit outbound actions that should remain distinct from the default reply target. 
async def node(state: State, runtime: Runtime[AionContext]) -> dict:
    thread = runtime.context.thread
    await thread.post("Posting a follow-up note to a different channel.")
    return {}
Unlike reply(...), post(...) should remain explicit in the response buffer so cross-target delivery is never inferred from transcript order alone.

​
history(...)

history(...) is intended to ask the control plane for more context than was included on the inbound turn. 
async def node(state: State, runtime: Runtime[AionContext]) -> dict:
    recent = await runtime.context.thread.history(limit=5)
    return {}
This should compile down to a control-plane request that uses the distribution metadata already attached to the current turn.

​
typing()

typing() is intended for stream-only live intents. 
async def node(state: State, runtime: Runtime[AionContext]) -> dict:
    await runtime.context.thread.typing()
    return {}
These intents should not replace the durable response precedence. They are best-effort live hints for streaming views.