t2t

Local push-to-talk dictation for macOS. Hold Fn to type. Hold Fn+Ctrl to talk to a local Pi agent.

Download

Download for macOS →

View all releases on GitHub →

t2t is currently unsigned. After replacing the app, macOS may require you to re-enable Accessibility permission for /Applications/t2t.app.

What it does

• Fn → record speech, transcribe locally with Whisper, paste into the focused text field.
• Fn+Ctrl → record speech, transcribe locally, send the prompt to the local pi CLI agent, then speak Pi's reply with macOS text-to-speech.
• Agent tools stay enabled. t2t delegates the agent loop to Pi rather than embedding a hosted chat client.
• Cloudflare AI Gateway is the recommended public model path. Users choose their Pi provider and model.

Requirements

• macOS
• Accessibility permission
• Microphone permission
• pi or a compatible pi CLI on PATH for Fn+Ctrl agent mode
• A Pi provider/model configuration

Recommended AI setup: Cloudflare AI Gateway

Pi already supports Cloudflare AI Gateway. Configure the provider outside t2t, then select it in t2t Settings:

export CLOUDFLARE_API_KEY=...
export CLOUDFLARE_ACCOUNT_ID=...
export CLOUDFLARE_GATEWAY_ID=...

pi --provider cloudflare-ai-gateway --model gpt-5.4-mini -p "hello"

In t2t → Settings, use:

• Pi binary: pi
• Provider: cloudflare-ai-gateway
• Model: any model available through your gateway

Cloudflare employee local mode

Jordan's local setup uses the existing Pi/OpenCode provider:

• Provider: opencode.cloudflare.dev
• Model: ember-alpha

That setup is local/private configuration, not required by the public repo.

Settings

The Settings tab includes:

• Theme
• Pi voice agent config
  • Pi binary
  • Provider
  • Model
• Speak agent responses aloud
• MCP server management / discovery surface
• History and analytics

History & logging

• Transcriptions are stored locally.
• Pi agent prompt/response summaries are stored locally.
• Logs: ~/Library/Logs/t2t.log
• Whisper model: ~/.cache/whisper/ggml-base.en.bin

Developer setup

cd desktop
bun install
bun run check
bunx tauri dev

Build:

cd desktop
bun run build

Tech stack

• Tauri + Rust
• Svelte 5 / SvelteKit
• whisper-rs local transcription
• local pi CLI agent bridge
• Cloudflare AI Gateway recommended for model routing
• macOS say for spoken agent replies

License

MIT