v0.1.2 — Per-app source picker + Generate buttons

What's new

Per-app audio source picker

• New menu in the recording view lets you scope the system-audio capture to a single running app (Brave, Chrome, Zoom, Teams, Slack, etc.) instead of the whole system. Useful when you only want the meeting audio and not your music or notifications.
• Built on SCContentFilter(display:including:exceptingWindows:) — no virtual audio device needed.

Generate buttons on the Transcript tab

• Three buttons live below "Copy transcript": Generate Notes, Generate Summary, Generate Action Items. Each switches to the relevant tab and runs the AI generation in-flight, so you don't have to hop tabs first.
• New AI-generated Notes with a structured prompt (Discussed / Decisions / Open questions / Next steps). Result lands in the Notes tab and is still hand-editable afterward.
• Notes tab labels the model + timestamp when AI-generated.

Build

• Signed with Developer ID and notarized + stapled by Apple notarytool.
• arm64-only macOS build.

Install

1. Download XooMe.dmg below.
2. Open it, drag XooMe to Applications.
3. First launch: grant Screen Recording permission (required for system audio capture). Microphone permission is optional but recommended.

Configuration

Open Settings (gear icon in sidebar) and add:

• Deepgram API key — required for transcription.
• Anthropic API key or a custom OpenAI-compatible endpoint — required for AI notes/summary/action items.

Pendant integration (Limitless/Omi) ships off by default; toggle it on in Settings if you have one.

v0.1.1 — Pendant opt-in, drag-drop upload, permission UX

What's new in v0.1.1

This is a course-correction release. Real-world use revealed the v0.1.0 build was over-tuned toward the pendant flow at the expense of the primary use case (capturing meetings at the Mac).

Pendant is now opt-in

The pendant integration is off by default. No stealth Bluetooth prompts for users who never asked for it.

• Enable in Settings → Pendant (optional)
• When enabled, real-time streaming over BLE is also off by default — the pendant just connects quietly and waits for you to click Sync now in Lifelog. (The "live transcribe over BLE" mode is mostly a debug feature; when you're at your Mac, the system-audio + mic capture already there is the better experience.)
• Pendant recordings live exclusively in the Lifelog view now. They no longer pollute the main Transcripts list.

Drag-and-drop upload (regression fix from Electron build)

• Drag any audio file (.m4a / .mp3 / .mp4 / .wav / .ogg / .flac / .webm / .aac) onto the Transcripts view to transcribe it via Deepgram's pre-recorded REST API
• "Upload audio" button in the Transcripts toolbar opens the standard macOS file picker; multi-select supported
• Uploaded files are copied into XooMe's recordings dir so the Detail view's audio player has a stable playback path

System-audio permission UX

The "mic captured but system audio was silent" failure mode (which hits whenever a fresh-bundle-ID install resets ScreenCaptureKit permissions) now surfaces clearly:

• RecordingController runs CGPreflightScreenCaptureAccess() before starting capture
• If permission is missing, recording bails immediately with a clear status message and opens System Settings → Privacy & Security → Screen Recording so you can grant access
• After granting, quit and relaunch (macOS requires this for Screen Recording to take effect)

Other

• Build script is unchanged; libopus still bundled via Frameworks/
• Apple-notarized DMG; Gatekeeper accepts on first launch with no quarantine prompt
• MIT licensed

Upgrade

Drop the DMG below into /Applications, replacing the prior install. Your existing transcripts at `~/Library/Application Support/XooMe/` are preserved.

If you had pendant integration working in v0.1.0, you'll need to re-enable it in Settings → Pendant (optional) since the default is now off.

v0.1.0 — First public release

XooMe v0.1.0

First public release. Native macOS app for audio capture, transcription, and AI-summarized memory bank — for meetings, calls, and the Limitless Pendant.

What's in the box

System audio + mic

• ScreenCaptureKit-based system audio capture (no BlackHole or virtual loopback driver needed)
• Mic mixing with per-source gain sliders + level meters
• Audio file recorded as 16 kHz mono AAC alongside the transcript

Pendant integration (Limitless / Omi)

• One-tap BLE pairing (TimeSync triggers the macOS pair dialog)
• Auto-reconnect when pendant comes back into range
• NAND sync — pulls stored audio pages, ACKs them so the device frees flash storage
• libopus decoder (Apple's built-in Opus produced static for CELT mode; libopus handles every mode)
• Time-gap encounter clustering in the Lifelog view (recordings within 5 min group into one card)
• Firmware-flag-driven session boundaries (`did_start_recording` / `did_stop_recording`)

Transcription + AI

• Deepgram WebSocket streaming for live transcription with diarization (Nova-2)
• Anthropic Claude for summary + structured action items (Haiku 4.5 / Sonnet 4.6 / Opus 4.7)
• Bring your own LLM via any OpenAI-compatible endpoint (Ollama, vLLM, OpenRouter)

Privacy

• All data stored locally in `~/Library/Application Support/XooMe/`
• One JSON per transcript, one M4A per recording — easy to back up, easy to inspect
• No telemetry, no analytics, no remote logging
• Apple-notarized, hardened runtime, signed with Developer ID

Install

Download `XooMe.dmg` below, mount it, drag XooMe.app into `/Applications/`. First launch will prompt for Microphone, Screen Recording, and (if you use a pendant) Bluetooth permissions — approve in System Settings → Privacy & Security and relaunch.

Open Settings (⌘,) and paste your Deepgram and Anthropic API keys.

Known limitations in this release

• Apple Silicon only in the prebuilt DMG. Build from source for Intel Macs.
• Encounter clustering is time-gap only — speaker / topic / setting clustering land in 0.2.x.
• No "Ask XooMe AI" yet — the cross-transcript chat memory bank is scaffolded but not wired up.
• Integrations are placeholders — Google Calendar, Obsidian, Cursor, etc. are stubbed in the UI.
• Audio playback in Detail view is a basic AVPlayerView. No transcript-synchronized scrubbing yet.
• Speaker rename UI isn't ported from the Electron prototype yet.

Built with

Swift 5.9 · SwiftUI + AppKit · ScreenCaptureKit · AVAudioEngine · CoreBluetooth · libopus · Deepgram · Anthropic Claude

License

MIT