Documentation

Learn how to get the most out of t2t

Getting Started

1. Download & Install

Download t2t from GitHub Releases. On first launch, you may need to right-click and select "Open" if macOS shows a security warning.

2. Grant Permissions

t2t needs two permissions:

  • Accessibility - Required to detect the Fn key and paste into the correct field
  • Microphone - Required for voice recording

The app will prompt you if permissions are missing. You can also grant them in System Settings.

3. First Run

On first launch, t2t automatically downloads the Whisper model (~150MB) to your cache directory. This happens in the background—you'll see a notification when it's ready.

Typing Mode

How to Use

  1. Focus any text field (email, notes, code editor, etc.)
  2. Hold the fn key
  3. Speak your text
  4. Release fn to transcribe and paste

Visual Feedback

A red bar appears at the top of your screen while recording:

  • Red bar = Recording (typing mode)
  • Amber bar = Processing transcription
  • Bar disappears when text is pasted

Tips

  • Works in any app—no special integration needed
  • Your clipboard is preserved (t2t saves and restores it)
  • Speak clearly and at a normal pace for best results
  • The model runs locally—your voice never leaves your computer

Agent Mode

How to Activate

Hold fn + ctrl (or fn + cmd on macOS) to enter agent mode.

A cyan bar appears while recording in agent mode.

What It Does

Agent mode uses AI to understand your voice command and execute actions:

  • With MCP servers: Connects to your configured MCP servers and uses their tools
  • Without MCP servers: Generates and executes AppleScript for macOS automation

Setup Required

Agent mode requires an OpenRouter API key:

  1. Get a free API key from openrouter.ai
  2. Open t2t settings (menu bar icon → View Settings)
  3. Add your OpenRouter API key in the Settings tab
  4. Optionally select your preferred AI model

Example Commands

  • "Open Slack"
  • "Create a new note in Obsidian"
  • "Search my database for users created this week"
  • "Send a notification saying meeting in 5 minutes"

MCP Servers

What Are MCP Servers?

MCP (Model Context Protocol) servers provide tools and capabilities that extend what the AI agent can do. Instead of hardcoding integrations, you connect to MCP servers that expose their own tools.

Why Use MCP?

  • Extensible - Connect to databases, APIs, file systems, or any MCP-compatible service
  • Unlimited possibilities - Each server adds new tools the agent can use
  • Multiple servers - Connect to as many as you want simultaneously
  • Local execution - All tool execution happens on your computer

How to Configure

  1. Open t2t settings (menu bar icon → View Settings)
  2. Go to the "Settings" tab
  3. Click "New MCP Server"
  4. Fill in the details:
    • Name - A friendly name for the server
    • Transport - stdio (for local commands) or HTTP/HTTPS (for remote servers)
    • Command/URL - The command to run (stdio) or URL to connect to (HTTP)
    • Args - Command-line arguments (for stdio transport)
  5. Toggle the server on to enable it

Common Examples

Database Server (stdio)

npx @modelcontextprotocol/server-postgres

File System Server (stdio)

npx @modelcontextprotocol/server-filesystem

Remote API Server (HTTP)

https://your-mcp-server.com/api

Transport Types

  • stdio - Spawns a local process (e.g., `npx` commands). Best for local tools.
  • HTTP/HTTPS - Connects to a remote server via HTTP. Best for remote APIs.

Use Cases

Voice Dictation

Use typing mode to quickly input text anywhere:

  • Writing emails, messages, or documents
  • Coding comments and documentation
  • Taking notes during meetings
  • Filling out forms

Automation with MCP

Connect MCP servers to automate complex workflows:

  • Query databases: "Show me all users created this week"
  • File operations: "Create a new file in my project with this content"
  • API interactions: "Fetch the latest data from my API"
  • Custom tools: Connect to any service that supports MCP

macOS Automation

Without MCP servers, agent mode uses AppleScript for macOS automation:

  • Opening applications
  • System notifications
  • Basic app control

Troubleshooting

Permissions Not Working

If t2t isn't detecting the Fn key or can't paste:

  1. Check System Settings → Privacy & Security → Accessibility
  2. Ensure t2t is enabled
  3. Restart t2t after granting permissions

Model Not Downloading

If the Whisper model isn't downloading:

  • Check your internet connection
  • Verify disk space (model is ~150MB)
  • Check logs at ~/Library/Logs/t2t.log

MCP Server Connection Failed

If an MCP server won't connect:

  • Verify the command/URL is correct
  • For stdio: Ensure the command is available in your PATH (e.g., `npx` is installed)
  • For HTTP: Check the URL is accessible
  • Check the server status in settings (red = error, yellow = loading, green = active)
  • Review error messages in the settings UI

Agent Mode Not Working

If agent mode doesn't respond:

  • Verify your OpenRouter API key is set in settings
  • Check that you're holding fn+ctrl (or fn+cmd on macOS)
  • Look for error notifications
  • Check logs at ~/Library/Logs/t2t.log