The fastest way to input text on macOS.
Hold a key, say what you mean, and watch it appear — anywhere, instantly.

---

Download

Grab the latest .dmg from Releases.

Requirements: macOS 15+ (Sequoia), Apple Silicon (M1 or later).

Nightly builds are published as the rolling nightly prerelease.

CI

This public repository hosts the macOS nightly build workflow so GitHub-hosted macOS minutes are available to the release pipeline. The workflow checks out the private source repositories at build time:

• ontypehq/OnType
• ontypehq/libfst
• ontypehq/itn

Required repository secrets:

• ONTYPE_SOURCE_TOKEN: token with read access to the private source repos.
• NOTARY_KEY_P8_BASE64: base64-encoded App Store Connect API key file (AuthKey_UPM4638DP7.p8 locally).
• OnType release env copied from the private app repo .env: ONTYPE_AUTH_BASE_URL, ONTYPE_AUTH_BASE_URL_DEV, ONTYPE_AUTH_BASE_URL_PROD, CONVEX_URL_DEV, CONVEX_URL_PROD, DEVELOPER_ID_CERT_BASE64, P12_PASSWORD, KEYCHAIN_PASSWORD, APPLE_ID, APPLE_ID_PWD, APPLE_TEAM_ID, NOTARY_KEY_ID, NOTARY_ISSUER_ID, CF_ACCOUNT_ID, CF_R2_BUCKET, CF_R2_ACCESS_KEY_ID, CF_R2_SECRET_ACCESS_KEY, CF_R2_ENDPOINT, CF_ASSETS_BASE_URL, SPARKLE_ED_PRIVATE_KEY, SENTRY_ORG, SENTRY_PROJECT, SENTRY_AUTH_TOKEN, SENTRY_TRACES_SAMPLE_RATE, VITE_PUBLIC_POSTHOG_PROJECT_TOKEN, and VITE_PUBLIC_POSTHOG_HOST.

Nightly builds intentionally do not run bun run app:release --publish; they produce arm64 notarized DMGs and attach them to GitHub Releases only. Multi-arch builds and Sparkle/R2 publishing should stay on a separate, explicitly approved workflow for formal releases.

How it works

Hold your hotkey → Speak naturally → Release to type. That's it.

OnType runs on-device AI to transcribe your voice in real time. Your audio never leaves your Mac.

Features

• On-device ASR — Powered by MLX on Apple Silicon. No cloud, no latency, no subscriptions.
• Works everywhere — Any text field in any app. Xcode, Slack, Notes, browsers, terminals.
• AI rewriting — Optionally clean up grammar, formatting, or tone after transcription.
• Privacy by default — 100% offline capable. Zero telemetry. Your voice data stays on your machine.

Feedback

Found a bug or have a feature request? Open an issue.

Links

• Website
• Changelog