Repolode
Submit
farshed

farshed/sage

A self-hosted voice chat app in Rust that lets you talk with LLMs using open-source speech recognition and text-to-speech.

View on GitHub
Stars
464
Forks
38
Open issues
2
Last commit
2 months ago

What is it?

Sage is a self-hosted voice chat application for talking with large language models. Speech is transcribed locally with Whisper, the model's reply is generated by an LLM, and the answer is spoken back using the Kokoro text-to-speech model, so the speech pipeline can run offline.

For text generation you can run a local model through Ollama or use a third-party provider. Deepseek, OpenAI, Anthropic and Together.ai are supported out of the box, and switching between them is done by setting environment variables in a .env file. Sage can run as a Docker container or natively, with setup scripts for Unix and Windows; the native build on macOS uses CoreML for Whisper.

The project is aimed at people who want a voice interface to an LLM without sending audio to a cloud speech service. The README notes that Docker inference with Whisper is slower than native, and lists CUDA support, multilingual support and a configurable system prompt as planned work.

Key features

  • 01Local speech-to-text with Whisper and text-to-speech with Kokoro.
  • 02Supports Ollama for self-hosted models and Deepseek, OpenAI, Anthropic and Together.ai as providers.
  • 03Provider and model are chosen through environment variables.
  • 04Runs in Docker or natively on Unix and Windows.
  • 05Uses CoreML for Whisper on macOS.

Use cases

  • Talking to a local Ollama model by voice without sending audio to a cloud service.
  • Adding a voice interface to an LLM provider you already use.
  • Experimenting with an offline speech-to-LLM-to-speech pipeline on consumer hardware.

In collections

Self-hosted appsRun LLMs locallyAI video & audio tools

Topics

#voice-assistant#llm#speech-to-text#text-to-speech#whisper#self-hosted#ollama#rust

FAQ

What is sage?
A self-hosted voice chat app in Rust that lets you talk with LLMs using open-source speech recognition and text-to-speech. Sage is a self-hosted voice chat application for talking with large language models. Speech is transcribed locally with Whisper, the model's reply is generated by an LLM, and the answer is spoken back using the Kokoro text-to-speech model, so the speech pipeline can run offline.
Is sage free to use? What license does it have?
sage is open source under the MIT license. Check the license terms before using it in commercial products.
Is sage still maintained?
Updated within 6 months. The last commit to sage was on Sep 16, 2026.
What can I use sage for?
Talking to a local Ollama model by voice without sending audio to a cloud service. Adding a voice interface to an LLM provider you already use. Experimenting with an offline speech-to-LLM-to-speech pipeline on consumer hardware.
Where can I get sage?
The source code and installation instructions are on GitHub: https://github.com/farshed/sage

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects