Sonia is a local voice assistant that combines an LLM with speech-to-text and text-to-speech and needs no dedicated GPU, running in less than 15 GB of RAM entirely on a local computer. It is multilingual and full-duplex, so the microphone stays open while the assistant talks and you can interrupt it and it pauses, resumes or cancels accordingly. The only external piece is an OpenAI-compatible endpoint.
Be the first to rate Sonia Assistant. Your feedback helps other users decide.
Everything runs locally in under 15 GB of RAM.
Full-duplex audio means barge-in pauses, resumes or cancels.
It talks to any OpenAI-compatible endpoint such as a running llama-server.
Similar tools in the same category โ compare side-by-side
Braus gives agents a desktop workspace on your Mac, so agents work in your apps while you keep using...
TinyWebUI is a small, local-first chat workspace for any OpenAI-compatible model, with web search an...
DeepSearch is an independent web search client for the DeepSeek API. Once you sign in it returns ran...
Guides and comparisons about productivity tools
Ten launches that put agents on hardware: a phone that drives itself, cloud Android devices, live captions on a Mac, and voice assistants with no chat box.
Discover 5 AI tools for customer support automation in 2026. Reduce response times with AI chatbots, ticket routing, and sentiment-based escalation.
Discover the best free AI tools for digital nomads in 2026. Manage timezones, track expenses, find co-working spaces, and stay productive on the road.
Discover the best free AI tools for freelancers in 2026. From proposals and invoicing to writing assistance and project management.
Discover the best free AI tools for startups in 2026. From pitch deck generation and financial modeling to investor matching and milestone tracking.