Ichigo

Ichigo is an open-source speech package from the Jan (janhq) team for building local, real-time voice AI, bundling speech recognition, a speech-aware language model and text-to-speech in one toolkit. It has around 2,500 GitHub stars.

Overview

Ichigo was designed for developers who want voice interfaces that run on their own hardware instead of streaming audio to a cloud API. The package is organised into three parts: Ichigo-ASR for transcribing speech, Ichigo-LLM — a Llama-3-based model that can listen and reason about audio — and Ichigo-TTS for speaking responses back. Together they cover the full loop of a voice assistant: hear, think, talk. Because everything is packaged for developers — Python APIs rather than a consumer app — it is meant to be embedded inside other products and experiments. The project comes out of the same community as the Jan desktop app and the earlier Homebrew work on giving Llama ears, and it includes a paper, benchmarks and hosted model weights. It was last updated in 2025.

Key Features

• Speech recognition, speech LLM and TTS in one package
• Runs locally for private, low-latency voice interactions
• Streaming-friendly design aimed at real-time conversation
• Llama-3-based model that understands spoken input directly
• Components can be used together or independently
• Part of the wider Jan open-source ecosystem, weights on Hugging Face

💳 Pricing

Ichigo is free and open source. You download the code and weights and run them on your own hardware, with no API fees; costs are limited to the GPU you run it on.

Who Is It For

Developers building voice assistants, accessibility tools and spoken interfaces who want the whole pipeline local and hackable, and researchers prototyping speech-to-speech systems.