Sonia Assistant

Sonia Assistant: Sonia is a local voice assistant that runs entirely on a computer without a dedicated GPU.

Sonia is an LLM, STT, and TTS voice assistant that operates entirely on a local computer. It requires no dedicated GPU and runs in less than 15 GB of RAM. The assistant features true full-duplex listening, allowing the microphone to stay open so you can interrupt replies mid-sentence. It is bilingual in English and Italian, automatically mirroring whichever language you speak. Supported inputs and outputs include streaming ASR, streaming TTS, and an OpenAI-compatible LLM endpoint. It includes function-calling tools such as datetime, timers, watches, notes, skills, and shell commands, managed through a browser app-window UI or terminal modes.

Sonia Assistant at a glance

Best for: Software engineer
Runs on: Web, Linux, CLI, self-hosted, open source
Pricing: Free
Released: October 2026
Category: AI agent tool
Website: github.com/grigio/sonia-assistant

Key features

True full-duplex listening with barge-in support
Streaming ASR and layered turn detection
Bilingual English and Italian localization
Function-calling tools for datetime, timers, and shell
Externally managed SKILL.md packages
Task banner over chat for timers and watches
Persistent chat and conversation rewind

Use it for

Manage timers and watches with voice commands
Run shell commands and scripts hands-free
Take notes and query datetime locally
Chat interactively in English or Italian

Alternatives to Sonia Assistant

Tech blog — Runs locally with open-microphone barge-in and no cloud dependency.
Phocinae-Largha-150M-v1（斑海豹 Largha） — Runs locally with zero generated tokens and high decision speed.
Leon — It keeps data private by running on your own server.
MimiClaw — Runs AI agents directly on low-cost hardware without an operating system.
Memori — Reduces LLM token usage and spend by over 97% with targeted recall.
Waku Agent — Local-first memory storage and readable Python code for AI agent logic.
py-xiaozhi — Connects LLMs to physical hardware and local IoT devices via async architecture.
Goose — Runs locally on your machine with support for 70+ extensions and any LLM.
Lokutor — Runs the full voice pipeline on standard CPUs for 4 to 5 cents a minute.
Tigerless Agent Memory — Maintains agent history in portable, greppable Markdown files without API keys.
Ming Voice — It runs fully local with low latency on your own GPU without sending audio elsewhere.
Beeuni — Agents retain context and improve skills over time through a controlled review process.

All Sonia Assistant alternatives