Privacy-first, local-first AI meeting assistant for macOS & Windows.
Meetwit listens to your live meetings, indexes your local company documents, and answers questions in real time using both — with sources. It detects conflicts between what's being discussed and what your company has already decided.
Everything runs locally on your machine. Nothing leaves. No accounts, no
telemetry, no cloud — the app talks only to localhost.
Most meeting assistants transcribe what was said. Meetwit also understands what your company has already decided, so it can:
- Index local company documents (PDF, DOCX, Markdown, TXT)
- Listen to live meetings (mic + system audio) and produce a real-time transcript
- Answer questions during the meeting using your docs + meeting history, with sources
- Detect conflicts between meeting content and your company knowledge
- Save summaries, decisions, and action items for cross-meeting search
- Live transcription — mic + system audio, on-device Whisper (GPU-accelerated: Metal on macOS, Vulkan on Windows)
- Multilingual — transcribe in any language; write summaries in any language, independent of the spoken one
- In-meeting copilot — ask questions during the call, grounded in your docs + transcript, with citations
- Live notes — jot timestamped notes while recording
- Summaries, decisions & action items — generated locally after the meeting
- Cross-meeting memory — semantic search across every meeting and document
- Conflict detection — flags when a new decision contradicts a past one
- Organize — nest meetings in folders, merge interrupted sessions
- Import audio — transcribe an existing recording
- Export — Markdown, PDF, plain text, WebVTT, SRT, JSON
- Bring your own model — local Ollama by default; optional OpenAI / Anthropic / Groq / OpenRouter (keys stay in the OS keystore — macOS Keychain / Windows Credential Manager)
By default, zero outbound network requests. The app talks only to:
- a random
127.0.0.1port — the auto-spawned local Python sidecar localhost:11434— your local Ollama install
No analytics, no crash reporting, no accounts. Your meetings, transcripts, and recordings never touch a network. There's no telemetry to opt out of — there's none to begin with, and you can verify that in the source.
Grab the latest build from the
Releases page and verify
your download against the published SHA-256.
macOS (Apple Silicon)
- Download
Meetwit_*_aarch64.dmg. - Open the
.dmgand drag Meetwit to your Applications folder. - Launch it and grant Microphone and Audio Capture permission — both are needed to capture your side of the call and the other participants' audio.
macOS builds are signed with a Developer ID and notarized by Apple, so the app opens normally — no "unidentified developer" warning.
Windows (x64)
- Download the
*-cpuinstaller (.exeor.msi) — it runs on any PC. If you have a Vulkan-capable GPU and want ~5–10× faster transcription, grab the*-vulkanbuild instead. - Run the installer. (Windows builds are not yet code-signed, so SmartScreen may warn — click More info → Run anyway.)
- Launch Meetwit and grant Microphone permission. System audio is captured via WASAPI loopback and needs no extra permission.
- macOS 13+ (Apple Silicon) or Windows 10/11 (x64)
- Ollama (for local LLM inference)
- ~3 GB free disk for the Whisper + embedding models
| Layer | Stack |
|---|---|
| Desktop shell | Tauri 2 + Rust + React 19 + TypeScript + Tailwind 4 |
| Speech-to-text | whisper-rs (Metal on macOS, Vulkan/CPU on Windows) |
| Audio | Core Audio tap (macOS) / WASAPI loopback (Windows) for system audio + cpal (mic) |
| Backend | FastAPI + SQLite + sqlite-vec (auto-spawned sidecar) |
| Embeddings | BGE-M3 (multilingual) |
| LLM | Ollama (local) or BYOK cloud providers |
Prerequisites: rustup, Node 22 (via nvm), uv, pnpm, plus the Whisper
build toolchain (cmake on macOS; LLVM + the Vulkan SDK on Windows for the GPU
build).
./scripts/bootstrap.sh
pnpm tauri:devRelease build — macOS .dmg, or Windows .msi + .exe:
./scripts/build-release.sh
# Windows GPU build: pnpm tauri:build -- --features gpu-vulkanIssues and pull requests are welcome. Please open an issue to discuss larger changes before starting.
MIT.