Repository navigation
v2.26.0
Release 2.26.0 finishes the host-owned voice calls (docs/plans/voice-architecture.md, stages 2–4):
- The bot's work reaches the call: the host pushes the bot's final replies, approvals waiting, failures and rate-limited progress to the device holding the call (a
callevent older clients ignore); the apps just hand each line to the voice. Missed lines replay on reconnect. - One engine state machine on iPhone, Android and desktop: connecting → ready → listening/speaking, plus reconnecting, interrupted and failed. Gemini/OpenAI drops reconnect (3 tries) with the call so far; phone calls, Siri, other apps' audio and route/device changes pause and resume without hanging up; a playback watchdog keeps the #214 shape from holding replies;
voice_call_qualityanalytics (no content). - Outside the app: the bot's Live Activity shows the call — avatar, call timer, bot name, and mute and end buttons that work without opening the app (
LiveActivityIntents routed to the call bar's own mute and end) — plus the bot's progress; Android's call notification shows the call state, a call timer and progress. CallKit was built and then removed before release by the owner's decision; the design and how to restore it are archived indocs/plans/callkit.md. - Language: the voice answers in the language the user is speaking right now and follows when they switch (the phone's language only covers speaking first).
minHost2.26.0 in the iPhone, Android and desktop apps (they rely on the host-pushed lines);MIN_APPunchanged, so hosts update before phones.
Refs #214
What's Changed
- feat: release 2.26.0 with calls that follow the bot's work, reconnect and show on the system by @leepokai in #219
Full Changelog: v2.25.0...v2.26.0