v0.12.55+12055-macos
BasedHardware/omiv0.12.55+12055-macosJul 9, 2026by mdmohsin7
AI Summary
Major desktop update focusing on screen-aware agent launches, push-to-talk improvements, and fixes for background agent continuity and main-thread performance.
Key Highlights
- Hardened screen-aware agent launches and tool error reporting
- Improved desktop chat screen-context detection and diagnostics
- Fixed main-thread freeze and CPU pegging on text selection
- Improved PTT screen grounding for voice turns
- Fixed background agent chat continuity after pills dismiss
New Features
- Background agent chat links reopen from timeline
- Typed notch and PTT turns appear in shared history
- Floating bar reads from shared chat timeline
- Polished desktop Home stage layout and wider chat surface
Full Release Notes
## OMI Desktop v0.12.55 ### What's New - Hardened screen-aware agent launches, tool schemas, and realtime tool error reporting. - Improved desktop chat screen-context detection and diagnostics - Background agent chat links now reopen from the timeline even after pills dismiss - Fixed Background agent cards so the header collapses again, markdown renders in the expanded body, and the open-agent link hides when the agent is unavailable - Fixed floating bar treating block-only or artifact answers as failed empty responses - Fixed silent agent-link opens and restored background-agent chat continuity after pills dismiss - Fixed a main-thread freeze where chat text selection wrapped agent card headers and pegged CPU - Show typed notch, notch PTT, and spawned-agent link turns in the shared Omi chat history - Background agent cards and proactive notifications now stay consistent across chat surfaces after reload - Fixed continuity review findings: restage can create a missing assistant after a user-only stage, owner-swap clear no longer clobbers a mid-session auth uid, agent open ordering is generation-guarded, and agent cards hide dead Open links / avoid false-success status icons - Floating bar chat now reads from the shared chat timeline instead of keeping a separate copy - Attach Gemini push-to-talk screen frames immediately after activity start instead of at commit so current-screen questions have time to ground on the image - Send Gemini push-to-talk screen frames inside the active voice turn before commit so screen questions ground on current pixels instead of a too-late screenshot tool call - Fixed Home chat input focus after expansion and added Escape-to-Home navigation on secondary Home pages - Polished the desktop Home stage layout, wider chat surface, responsive ask bar, and icon-only notch thinking state - Fixed notch and chat so agent artifacts stay attached to the turn that produced them, old orphan artifact cards no longer appear in the floating viewport, and agent list/card previews show the prompt instead of the response - Fixed a test-owner swap that could wipe sign-in tokens and break chat loading until relaunch - Make push-to-talk screen context fresher and less brittle by prefetching canonical voice-turn screen context while the user speaks and preventing stale screenshot OCR from biasing transcript correction - Made push-to-talk screen context prefer live capture when finalized screen frames are more than a few seconds old - PTT now tries available Omi context before asking follow-up questions and delegates larger work to background agents when a short voice answer is not the right surface - Fixed duplicate chat messages after push-to-talk - Buffered push-to-talk screen context until realtime sessions are ready so cold starts and reconnects do not drop it - Fixed subagent chats so completed agents show their final output and stop showing stale Running or Working indicators - Keep main chat subagent links visible without showing completed implementation-only tool logs, and keep externally spawned agent pills synced to terminal runtime state ### Downloads - **DMG Installer**: For fresh installs, download the DMG below - **Auto-Update**: Existing users will receive this update automatically via Sparkle <!-- KEY_VALUE_START isLive: true channel: beta edSignature: PA6w5nM1N/xdim3ICUdB6tmm4VWk9lVbBNdHofnQv2ADMqcCbCYco7WE3maKKLHGPAxJMAXMI4JQes7lcnHoBg== changelog: Hardened screen-aware agent launches, tool schemas, and realtime tool error reporting.|Improved desktop chat screen-context detection and diagnostics|Background agent chat links now reopen from the timeline even after pills dismiss|Fixed Background agent cards so the header collapses again, markdown renders in the expanded body, and the open-agent link hides when the agent is unavailable|Fixed floating bar treating block-only or artifact answers as failed empty responses|Fixed silent agent-link opens and restored background-agent chat continuity after pills dismiss|Fixed a main-thread freeze where chat text selection wrapped agent card headers and pegged CPU|Show typed notch, notch PTT, and spawned-agent link turns in the shared Omi chat history|Background agent cards and proactive notifications now stay consistent across chat surfaces after reload|Fixed continuity review findings: restage can create a missing assistant after a user-only stage, owner-swap clear no longer clobbers a mid-session auth uid, agent open ordering is generation-guarded, and agent cards hide dead Open links / avoid false-success status icons|Floating bar chat now reads from the shared chat timeline instead of keeping a separate copy|Attach Gemini push-to-talk screen frames immediately after activity start instead of at commit so current-screen questions have time to ground on the image|Send Gemini push-to-talk screen frames inside the active voice turn before commit so screen questions ground on current pixels instead of a too-late screenshot tool call|Fixed Home chat input focus after expansion and added Escape-to-Home navigation on secondary Home pages|Polished the desktop Home stage layout, wider chat surface, responsive ask bar, and icon-only notch thinking state|Fixed notch and chat so agent artifacts stay attached to the turn that produced them, old orphan artifact cards no longer appear in the floating viewport, and agent list/card previews show the prompt instead of the response|Fixed a test-owner swap that could wipe sign-in tokens and break chat loading until relaunch|Make push-to-talk screen context fresher and less brittle by prefetching canonical voice-turn screen context while the user speaks and preventing stale screenshot OCR from biasing transcript correction|Made push-to-talk screen context prefer live capture when finalized screen frames are more than a few seconds old|PTT now tries available Omi context before asking follow-up questions and delegates larger work to background agents when a short voice answer is not the right surface|Fixed duplicate chat messages after push-to-talk|Buffered push-to-talk screen context until realtime sessions are ready so cold starts and reconnects do not drop it|Fixed subagent chats so completed agents show their final output and stop showing stale Running or Working indicators|Keep main chat subagent links visible without showing completed implementation-only tool logs, and keep externally spawned agent pills synced to terminal runtime state KEY_VALUE_END -->