First launch now opens a consent dialog. We collect app launches, model usage, and error reports — never your audio, transcribed text, or personal info. Existing users are grandfathered; flip the toggle in the app’s Settings any time, no account needed.
v2.13💾
Smoother first-run install
Redesigned DMG with a clear “Drag to Applications” arrow, plus a friendly prompt if you accidentally launch the app straight from the disk image. Apps that live in /Applications behave correctly — this just makes sure yours ends up there.
v2.11🔇
Mute system audio while you record
YouTube, Spotify, a Zoom call on speakers — quietly ducked while a recording is in progress, then restored. No more transcripts polluted by whatever was playing in the background.
v2.10🧹
Per-model hallucination filtering
Whisper, Parakeet, and Cohere each have their own characteristic filler outputs. SpeechButton now ships a tuned word-list per model, so “you”, “thanks for watching”, and the rest don’t end up in your text.
v2.10📋
Hands-free copies to clipboard
Hands-free chunks now land on the clipboard instead of being pasted into whatever window has focus. Speak, then paste where you actually want the text.
v2.9🎤
Notch-aware recording indicator
A pill toast that wraps around the MacBook notch with a floating pulse during recording. Configurable position, sized properly for external displays.
Voice-to-agent in milliseconds. Everything runs on your Mac.
🔒
100% Private & Local
Your voice never leaves your Mac. Speech recognition runs on the Apple Neural Engine via CoreML. AI transforms also run locally. No internet required.
🌍
Multi-Language Support
Multiple speech models to choose from. Fast English recognition or 100+ languages. Auto-detect or set your language explicitly.
⚡
20ms Capture — Never Lose the First Word
Press the key — recording starts in 20 milliseconds. Other tools take 200ms+ and clip your first word. When you're rapid-firing tasks to 3 agents, every lost word means repeating yourself.
🛠
Config as Code — Agents Configure Agents
Plain TOML config + shell scripts. Your AI agent can add channels, write transforms, and configure routing — programmatically. The tool evolves with your workflow.
🔌
Per-Hotkey Agent Routing
⌘+1 sends to Claude Code. ⌘+2 sends to Slack. ⌘+3 creates a Linear issue. Each hotkey is a programmable channel with its own transform pipeline and destination. No other tool does this.
✨
AI Transform Pipeline
Your rambling speech becomes a structured task. Transform scripts turn "fix the auth bug where tokens expire" into a formatted Problem/Steps/Priority prompt that your agent acts on immediately.
Industry First
Fully Local AI Pipeline
Voice recognition and AI text transform — both running on your Mac. Zero data leaves your device.
🎤
Your Voice
→
↓
Speech-to-Text
On-device · CoreML
→
↓
AI Transform
Local · prompt-driven
→
↓
📄
Polished Text
Pasted at cursor
Nothing leaves your Mac. Ever.
Trusted by developers and creators
🤖Claude Code
💬Slack
✅Linear
💻VS Code
📝Notion
🖥Terminal
★★★★★
"With other apps I kept losing the first word and had to repeat myself. SpeechButton captures everything from the first syllable. Saved me 2 hours a day on documentation alone."
Marcus Torres Senior Software Engineer, Stripe
★★★★★
"I walk around the room and code by voice. My back pain is gone. Wrote 40% more code last month just by dictating instead of typing. Development feels completely different now."
James Kim Full-Stack Developer, Linear
★★★★★
"I manage 5 AI agents just by talking. Hold a key, give a task, release. Cut context-switching time by 70%. The channel routing is a game changer for multi-agent workflows."
SpeechButton transcribes in chunks as you speak. Every time you pause briefly, the chunk is transcribed and pasted instantly. When you finish talking, almost all text is already there.
With auto-Enter enabled, a longer silence (3s) sends the full message automatically. Hold the hotkey, talk to Claude Code or ChatGPT, pause briefly between thoughts — chunks appear in realtime. Stop talking for 3 seconds — message is sent. No keyboard needed.
"necesito revisar el servidor antes de la reunión de mañana"
✨ Prompt
"Translate to English. Keep it natural."
🚀 Delivered
I need to check the server before tomorrow's meeting.
→ pasted in Slack
Voice Routing
Route Voice to Your Entire Dev Stack
Text doesn't just go into your active app. Route your speech to AI agents, work tools, social media, and more — each with its own hotkey channel.
🤖
Claude Code
Remote agent tasks
💻
Cursor / Windsurf
AI coding assistants
🖥
Terminal
Run commands by voice
✅
Linear / GitHub
Issues & PRs
💬
Slack / Discord
Team updates
📝
Notion / Obsidian
Notes & docs
🛠
Git
Commit messages
🚀
Webhooks
Zapier, Make, custom APIs
📧
Email
Send via API
📦
Telegram
Bots & channels
📰
Social Media
Twitter, LinkedIn
🔗
Any API
curl, scripts, you name it
Each destination is a [[hotkey]] channel in your config. Press ⌘+1, ⌘+2, ⌘+3 — each routes to a different place.
Built for Developers Who Ship by Voice
Hands-free agent workflows, voice-driven automation, and programmable text pipelines.
🗣
Hands-Free with VAD + Auto-Enter
Voice Activity Detection sends text as you speak. Combined with auto-Enter, you can talk to AI agents (Claude Code, ChatGPT, Slack bots) without touching the keyboard at all.
🛠
Text Config File
All settings in a single config.toml file. AI agents can configure SpeechButton programmatically — no GUI needed. Changes apply instantly without restart.
✨
Text Transform Pipeline
Process text before it's pasted: run it through a script, send it to an LLM API, or transform it locally. Get cleaned-up, formatted, or translated text — all from your voice.
🔌
Hotkey Channels
Hold Command, then press 1, 2, or 3 to route your speech to different destinations. Send voice to one agent, then switch to another with a different channel — perfect for multi-agent workflows.
🔗
Webhook & File Output
Send transcribed text to a webhook URL for integrations, or log everything to a file for history and audit. All outputs work simultaneously — paste, file, webhook, and exec at once.
📱
iPhone as Wireless Mic
Use your iPhone as an external microphone. With keep_hot = true the mic stays always-on — no 300ms wake-up delay when you start talking. Same blazing fast response, even over wireless.
{}
Text & JSON Output
Choose output_format = "text" for plain text or "json" for structured data with timestamps, language, and confidence — ideal for programmatic pipelines.
~/.config/speechbutton/config.toml
[global]
model = "parakeet-tdt-0.6b-v3-int8"# Default: paste transcription at cursor
[[hotkey]]
key = "RightCommand"
name = "default"
paste = "accessibility"# ⌘+1: clean up speech with Local AI (free, offline)
[[hotkey]]
key = "RightCommand"
channel = "1"
name = "cleanup"
transform = "prompts/cleanup.md"# ⌘+2: translate to English with Local AI
[[hotkey]]
key = "RightCommand"
channel = "2"
name = "translate"
transform = "prompts/translate_en.md"# ⌘+3: send to Slack
[[hotkey]]
key = "RightCommand"
channel = "3"
name = "slack"
transform = "prompts/slack_message.md"
exec = "SLACK_WEBHOOK_URL=https://hooks.slack.com/xxx integrations/send_slack.py"# ⌘+4: create Linear issue
[[hotkey]]
key = "RightCommand"
channel = "4"
name = "linear"
transform = "prompts/linear_issue.md"
exec = "LINEAR_API_KEY=lin_api_xxx integrations/send_linear.py"# Auto-detect iPhone mic, keep stream hot
[[device_rule]]
match = "iPhone"
keep_hot = true
Frequently Asked Questions
A push-to-talk voice router for macOS. Hold a hotkey, speak a task, release — your AI agent executes it. Different hotkeys route to different agents, each with its own transform pipeline. Think of it as a voice-first interface for your entire dev stack.
Vibe coding is writing software by describing what you want in natural language and letting AI agents write the code. SpeechButton makes this hands-free: you speak the task, the agent codes it. No typing, no copy-pasting prompts.
20ms capture latency (vs ~500ms for macOS Dictation), per-hotkey channel routing to send voice to different apps, a text transform pipeline to clean up or translate speech before pasting, and 100% local processing. macOS Dictation can't do any of that.
Yes. The free tier gives you 5 minutes of transcription per day. No account needed, no credit card. If you need unlimited transcription, Pro is $7.99/month.
No. All processing runs locally on the Apple Neural Engine via CoreML. No internet connection is required for transcription. Your voice data never leaves your Mac.
Multiple speech models: fast English recognition or 100+ languages. Language detection is automatic — just speak naturally.
Currently Apple Silicon only (M1 and later). Requires macOS 14 Sonoma or later. The Neural Engine on Apple Silicon is what makes the 20ms capture latency possible.
Assign different hotkeys (Cmd+1, Cmd+2, etc.) to different destinations. Each hotkey becomes a "channel" that routes your voice to a specific app or service. Perfect for multi-agent workflows: ⌘+1 sends a refactoring task to one Claude Code agent, ⌘+2 sends a test-writing task to another, ⌘+3 updates your team on Slack. Three agents, three hotkeys, zero typing.
Process your speech through a script or LLM before pasting. Clean up filler words, translate to another language, format as commands, or run any custom transformation. Your voice goes in, polished text comes out.
Yes. Your iPhone works as a wireless microphone with zero extra setup via Apple Continuity. SpeechButton automatically detects it and you can even set keep_hot = true to keep the mic always-on for instant response.
Cancel anytime through the Mac App Store. Go to Settings → Apple ID → Subscriptions → SpeechButton → Cancel. No questions asked, no hoops to jump through.
Yes! SpeechButton has a full GUI where you can configure everything — hotkeys, channels, transforms, VAD settings. But we also offer a plain-text config.toml file, which is rare for dictation apps. This means an AI agent can configure SpeechButton for you. For powerful routing setups, we recommend using small Python scripts — just tell your agent where you want to send messages, and it will write ~40 lines of code that gives you a seriously powerful voice pipeline.
Start Shipping by Voice
Your AI agents are waiting for instructions. Stop typing them.
