Wave is a lightweight, native macOS dictation app focused on fast voice-to-text workflows with minimal UI overhead. Press a shortcut, speak, and your words are instantly pasted at the cursor. Supports on-device transcription via Whisper and cloud transcription via Groq, plus an AI Mode that sends your voice to an LLM and pastes the response directly.

• Dictation — global shortcut triggers recording; transcription is pasted at the active cursor
• Push to Talk or Toggle — hold to record and release, or press once to start and again to stop
• Local transcription — on-device Whisper inference, no internet required
• Groq cloud transcription — faster cloud-based transcription via the Groq API
• AI Mode — separate shortcut sends your voice to an LLM and pastes a direct answer
• Snippets — save reusable text snippets; AI Mode is aware of them
• Dictation history — recent transcriptions with right-click copy
• Language selection — auto-detect or set a specific language (ISO 639-1)
• Custom vocabulary — bias the model toward specific words and names
• Microphone selection — choose any input device or use the system default

Both shortcuts are fully customizable in Settings → Shortcut.

Download the latest DMG from Releases.

Releases are distributed as signed, notarized DMGs.

make build

open build/Build/Products/Release/Wave.app

Or launch from Xcode — open Wave.xcodeproj, select the Wave scheme, and run.

GitHub Actions builds signed and notarized release DMGs for tags matching v*.*.*.

Required repository secrets:

• BUILD_CERTIFICATE_BASE64 — base64-encoded Developer ID Application .p12
• P12_PASSWORD — password for the .p12
• KEYCHAIN_PASSWORD — temporary CI keychain password
• APPLE_ID — Apple ID email used for notarization
• APPLE_APP_SPECIFIC_PASSWORD — app-specific password for notarization
• APPLE_TEAM_ID — Apple Developer Team ID
• SPARKLE_PRIVATE_KEY — Sparkle EdDSA private key for appcast generation

We also maintain a self-distributed "Direct" channel (Developer ID signed + notarized DMGs, Sparkle updates) hosted entirely on Cloudflare, following the same pattern used for the Ayron app:

• Artifacts (DMG + appcast.xml): published to a Cloudflare R2 bucket (wave-updates or equivalent) via scripts/release/publish-r2.sh.
• Custom domain: updates.wave.mxv.sh (or similar) serves appcast.xml and /downloads/Wave-latest.dmg (stable name, short cache) + versioned DMGs (long cache).
• Feed URL: https://updates.wave.mxv.sh/appcast.xml (see Config/Info-Direct.plist and build settings; the public Ed key lives here too).
• Landing page (Astro site at wave.mxv.sh, deployed with wrangler): prefers the direct Wave-latest.dmg; falls back to GitHub release assets. This is the "site" piece — update landing/src/data/site.ts (directDmgUrl / directAppcastUrl) and the DownloadModal as needed. (The Ayron marketing/docs sites were built in their "Verso"/Nextra setups; here we extended the existing Astro landing.)

1. One-time machine setup (release Mac):

   • brew install create-dmg awscli
   • Store notary profile: xcrun notarytool store-credentials "wave-notary" ...
   • Developer ID provisioning profile named "Wave Self Distribution" (or adjust DEVELOPER_ID_PROFILE_NAME).
   • Sparkle EdDSA key: run Sparkle's generate_keys once; private goes to Keychain, public to the SUPublicEDKey in the Info plist(s).
   • Copy scripts/release/wave-release.env.example → ~/.pi/agent/wave-release.env (or $WAVE_RELEASE_ENV) and fill real R2 creds + endpoint. The bucket + custom domain must be configured in Cloudflare (R2 + DNS for the updates subdomain).

2. Build the DMG (notarizes, staples, signs with Developer ID):

   make release-dmg            # reads version from Info(-Direct).plist
   make release-dmg 0.5.0 7    # or override
   

3. Generate the signed appcast (rewrites enclosure to stable -latest.dmg):

   make release-appcast
   

4. Publish to R2:

   source scripts/release/load-release-env.sh
   make publish-r2
   # or directly: scripts/release/publish-r2.sh
   

Combined target (after sourcing env) and other helpers are in the Makefile. See the scripts themselves (heavily commented, referencing the Ayron equivalents) and ExportOptions-Direct.plist + Config/Info-Direct.plist.

The old GitHub-pages appcast (mxvsh.github.io) and pure GH release DMG flow can coexist during transition; flip the SUFeedURL and landing primary URL when ready to cut over.

See also: scripts/release/*.sh, Makefile (release-* targets), Ayron's scripts/release/ for the reference implementation.

• Toggle and Push to Talk recording modes
• Local offline transcription with Whisper
• Groq cloud transcription
• AI Mode with LLM response via Groq
• Custom dictionary / vocabulary
• Language selection
• Dictation history with copy
• Snippets with AI awareness
• Microphone selection
• App-specific behavior profiles
• Quality presets for speed vs accuracy

Join the Discord community for feedback and help. For bug reports and feature requests, open a GitHub issue.

• whisper.cpp — local speech-to-text inference
• Sparkle — macOS auto-update framework
• PhosphorSwift — icon library

Contributions are welcome. See CONTRIBUTING.md for local setup and expectations.