# Izwi

A local-first voice AI runtime providing transcription, speaker diarization, text-to-speech, and voice cloning through a desktop app, web UI, CLI, and an OpenAI-compatible local API.

Izwi is a desktop app, web UI, CLI, and local inference server for voice AI. It runs entirely on the user's machine and exposes both product workflows and OpenAI-compatible API routes without requiring cloud services or API keys.

Capabilities include real-time voice conversations combining local speech recognition, chat, and text-to-speech; long-form text-to-speech, voice cloning, voice design, and saved voices; transcription, speaker diarization, and forced alignment; and local chat with model download, load, unload, and delete controls. OpenAI-compatible routes cover models, chat completions, audio speech, and audio transcriptions, with preview support for the Responses API. Supported model families include several text-to-speech and speech-recognition models.

It installs via a downloadable build for macOS, Linux, or Windows, is written in Rust, and is MIT licensed. Optional anonymous desktop analytics never include prompts, transcripts, audio, paths, or personal identifiers, and are disabled unless a user opts in.

Website: https://docs.izwiai.com/