v1.3.92026-04-20
Fixed keyboard focus loss while typing during AI replies, and polished the input card border on Mac.
What's New
- Fixed an issue where the input field would lose keyboard focus while you were typing, especially when another session was receiving an AI reply in the background.
- On Mac, the message input card now has a clearer, slightly stronger outline so it stands out better against the background.
v1.3.82026-04-18
Voice input works for long recordings, multi-device chat shows your messages instantly on every device, and the TUI shutdown feels less frozen.
Improvements
- Voice input now keeps transcribing during long recordings instead of stopping around the 1-minute mark.
- When you send a message from one device, it now appears on your other connected devices immediately — no more waiting for the AI reply before the bubble shows up.
- The desktop TUI now shows a "shutting down…" state while it's stopping background services, so the screen doesn't appear frozen during a slow stop.
v1.3.72026-04-18
Stop button now cleanly ends the AI session, Compact actually reduces tokens, and you can keep 10 active chats warm.
Improvements
- The concurrent-chat cap grew from 6 to 10, so switching between more sessions keeps each one warm instead of reloading.
Bug Fixes
- Tapping Stop now cleanly ends the current AI session for every provider (previously Codex, Copilot, OpenCode, and Gemini kept the underlying process running).
- Fixed Compact so it actually shrinks your conversation's token footprint. Before, the next message could reload the full pre-compact history from the provider, making Compact a no-op or even increase costs.
v1.3.62026-04-18
Gemini also runs as a long-lived chat process now, and follow-up messages sent during a running turn feel more natural.
Improvements
- Gemini chats now share the same persistent-process model as the other AI providers, so follow-up messages keep full context and cost less on long conversations.
- When you send a new message while the AI is still working, the current provider's native behavior handles it — Claude queues your message, while Codex, Copilot, OpenCode, and Gemini stop the current reply quickly so the new one can start without restarting the underlying process.
Bug Fixes
- Fixed a case where sending "stop" (or any follow-up message) used to re-send your earlier request after the "stop" response, making it look like the AI ignored you. The AI now responds directly to whatever you just sent.
v1.3.52026-04-18
Your recent AI chats stay warm in a reusable background pool.
Improvements
- AI chats now share a reusable background process pool: your most recent conversations stay warm so switching between them is instant, and chats left idle are automatically suspended in the background and ready to resume the moment you return.