🚀 New Features

3-Day Battery Life (firmware v3.0.18)
Omi devices now last up to 3 days on a single charge with the latest firmware. Update from device settings to opt in. A confirmation dialog walks you through what's changing before the flash starts.

Voice Replies on Mobile
Omi can now speak its responses back to you. Toggle Speak Omi's response aloud in Profile → Settings, with three modes: Off, Headphones-only (default), or Always. Voice replies also fire automatically when you press-to-talk on the device button — ask a question, get a spoken answer streamed back chunk-by-chunk.

Desktop: Omi Now Executes Tasks
Omi desktop can now take action on your behalf — running Omi tools end-to-end (file ops, captures, semantic search, screen capture, and more) inside the new agent runtime. (See video.)

Charging Status in App
The home screen battery indicator now shows a charging bolt when the device is plugged in, and the device detail page displays a dedicated Charging label and icon — translated across all 49 supported languages.

✨ Other Improvements

• Desktop floating bar: entire notification surface is now tappable, not just the icon
• Desktop floating bar: transcription quality substantially improved
• Desktop floating bar: new voice picker in settings + right-click Disable for 2 hours snooze
• Desktop beta now points to the dev backend so we can ship faster — expect occasional instability; please report anything broken
• Bring Your Own Keys (BYOK): Deepgram live + pre-recorded transcription, plus chat batch transcribe, now route through your own key when set, and bypass the freemium cap
• Phone calls: free-tier remaining-calls banner, +1 dialing now correctly maps to both US and Canada, and per-tier monthly quota tracking
• Action items: semantic search across all action items, vectorized at write-time and during conversation processing
• Battery history chart added to device diagnostics with day/week views, persisted locally on iOS and Android
• Onboarding: Bring Your Own Keys step now appears before Tasks; Skip on the speech-profile step works deterministically
• Desktop: Omi AI rebrand for the built-in assistant (previously pi-mono), with refreshed attribution
• Desktop: Claude 4.6 across all chat completion routes
• Desktop: capture-screen tool added to the Omi tools relay (now 14 tools)
• Daily reflection auto-notifications removed from the notification service
• 15+ new locale strings for charging, voice-response, phone-call, and consent — full coverage across all 49 supported languages

🐞 Bug & Performance Fixes

• Mobile: disabled the noisy 30-second device-disconnect notification (and the notify-on-kill registration) — disconnects no longer pile up in your notification tray
• Mobile: fix navigator pop crash on device disconnect
• Mobile: reset audio session on TTS playback failure to allow retry; activate session before first chunk
• Backend: dedicated TTS httpx client with keep-alive disabled (fixes connection-reuse hangs)
• Backend: action-items prompt restored to pre-Dec 28 quality bar after a regression
• Backend: daily recap no longer pulls action items from unrelated conversations
• Desktop: floating bar provider-switch race condition fixed (no more double bridges or stuck states)
• Desktop: chat completions request body limit raised to 16 MB (fixes large-screenshot failures)
• Desktop: prevent double tool execution in pi-mono mode
• Desktop: macOS system banner suppressed for proactive notifications (was duplicating the floating bar)
• Desktop: run.sh consolidated and made resilient to paths with spaces
• Backend: post-call cleanup and quota gating for phone calls

👥 Team

• New maintainer onboarded focused on stability and crash reduction. We're actively looking for more — if you want to help on the firmware, mobile, or desktop side, reply here or DM the team.

📌 Heads-Up

• Desktop beta is now pointed at the dev backend. Expect occasional instability and please file anything you hit
• Firmware v3.0.18 requires acknowledging the warning dialog before flashing — don't unplug during update