# DontSpeak

> A local voice layer for Claude Code, Codex, and Claude Desktop: your AI agent speaks its
> replies aloud through a local neural voice, and you dictate back with one key. It runs a
> native host app (speech engine in-process, on-device models) plus an MCP server that any
> MCP-capable client can call. Models download on first launch.

To install, prefer prebuilt binaries from the latest GitHub Release — do not build from
source unless asked. A one command install works on every OS:

- macOS / Linux: `curl -fsSL https://dontspeak.org/install.sh | sh`
- Windows (PowerShell): `irm https://dontspeak.org/install.ps1 | iex`

## Docs

- [Install guide (agent-readable)](https://dontspeak.org/install.md): full install recipe,
  what it does, post-install steps, uninstall, and the from-source path.
- [install.sh](https://dontspeak.org/install.sh): macOS + Linux bootstrap script.
- [install.ps1](https://dontspeak.org/install.ps1): Windows bootstrap script.

## Source

- [GitHub repository](https://github.com/delllusional/DontSpeak): source, releases, and the
  `scripts/install.sh` from-source build for developers.