Reusable Realtime Session across Handoffs & Agent Tasks

[toubatbrian]

No description provided.

[changeset-bot]

🦋 Changeset detected

Latest commit: dd61a2a

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 22 packages

Name	Type
@livekit/agents	Patch
@livekit/agents-plugin-google	Patch
@livekit/agents-plugin-openai	Patch
@livekit/agents-plugin-phonic	Patch
@livekit/agents-plugin-anam	Patch
@livekit/agents-plugin-baseten	Patch
@livekit/agents-plugin-bey	Patch
@livekit/agents-plugin-cartesia	Patch
@livekit/agents-plugin-deepgram	Patch
@livekit/agents-plugin-elevenlabs	Patch
@livekit/agents-plugin-hedra	Patch
@livekit/agents-plugin-inworld	Patch
@livekit/agents-plugin-lemonslice	Patch
@livekit/agents-plugin-livekit	Patch
@livekit/agents-plugin-neuphonic	Patch
@livekit/agents-plugin-resemble	Patch
@livekit/agents-plugin-rime	Patch
@livekit/agents-plugin-sarvam	Patch
@livekit/agents-plugin-silero	Patch
@livekit/agents-plugins-test	Patch
@livekit/agents-plugin-trugen	Patch
@livekit/agents-plugin-xai	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

💡 Codex Review

agents-js/agents/src/utils.ts

Lines 960 to 963 in 58726f4

	if (signal) {
	const abortPromise = waitForAbort(signal);
	while (true) {
	const result = await Promise.race([reader.read(), abortPromise]);

Handle pre-aborted signals before entering read loop

readStream races reader.read() against waitForAbort(signal), but waitForAbort only resolves on a future abort event. If the signal is already aborted when readStream starts (which can happen when cancellation wins a task startup race during pause/close), the abort branch never resolves and the generator keeps waiting on stream reads instead of exiting promptly, so cancellation can hang until the source ends.

---

agents-js/agents/src/voice/agent_activity.ts

Lines 829 to 830 in 58726f4

	taskFn: (abortController: AbortController) =>
	this.realtimeSayTask(handle, text, {}, abortController),

Preserve addToChatCtx behavior in realtime say path

The new realtime say branch drops the addToChatCtx option by calling realtimeSayTask without passing it. Unlike the TTS path, realtime generation always inserts an assistant message into chat context, so session.say(..., { addToChatCtx: false }) no longer works for realtime-without-TTS configurations and will unexpectedly mutate conversation history.

---

agents-js/agents/src/stream/deferred_stream.ts

Lines 87 to 89 in 58726f4

	if (this.detachRequested) {
	this.detachRequested = false;
	return;

Make detach state pump-local during source handoff

detachRequested is shared across all pump instances. During rapid detach+reattach, a newly attached pump can reach finally first (for example, if the new source ends immediately), clear this flag, and then the old detached pump will run the normal close path and close the shared writable side. That can terminate the newly attached stream unexpectedly and drop post-handoff audio.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[devin-ai-integration]

Devin Review found 2 new potential issues.

View 13 additional findings in Devin Review.

[devin-ai-integration]

🟡 realtimeSayTask error handler unconditionally sets agent state to 'listening' ignoring queued speeches

When this.realtimeSession.say() throws in realtimeSayTask, the catch block at line 2892 unconditionally calls this.agentSession._updateAgentState('listening'). This doesn't check whether there are other speeches queued or currently playing, unlike the established pattern in onPipelineReplyDone (agents/src/voice/agent_activity.ts:1577-1581) which checks !this.speechQueue.peek() && (!this._currentSpeech || this._currentSpeech.done()) before transitioning to 'listening'.

If say() is called and fails while other speeches are queued (e.g., from a concurrent generateReply), the agent state would briefly and incorrectly transition to 'listening' before the next queued speech corrects it to 'speaking'.

Suggested change

	} catch (e) {
	this.logger.error('failed to say text: %s', String(e));
	this.agentSession._updateAgentState('listening');
	return;
	} catch (e) {
	this.logger.error('failed to say text: %s', String(e));
	this.onPipelineReplyDone();
	return;

---

Was this helpful? React with 👍 or 👎 to provide feedback.