Verbi is an open-source, modular voice assistant application for experimenting with state-of-the-art speech transcription, language models and text-to-speech. Created by PromtEngineer, the GitHub project has around 1,100 stars and is actively maintained, with recent updates in late 2025.
Verbi is built like a workbench for voice AI: each stage of the voice pipeline — wake and capture, speech-to-text, response generation and speech synthesis — is a swappable module. You can plug in cloud APIs such as OpenAI, Groq, ElevenLabs, Cartesia and Deepgram, or keep everything local with Whisper-style transcription and models served through Ollama. That modularity makes it easy to benchmark providers against each other, prototype a voice product, or build a fully private assistant that never leaves your machine. The codebase is aimed at researchers and developers rather than end consumers: configuration is explicit, components are documented, and swapping one provider for another does not require rewriting the application.
The project README walks through choosing providers for each pipeline stage and wiring them together, and the maintainer actively documents experiments with new speech models. It is a practical starting point for comparing the latency, quality and cost of today’s voice APIs in one harness.
Verbi itself is free and open source under the MIT license. Cloud speech and LLM providers you connect may charge for usage under your own accounts; a fully local configuration runs at no cost.
Voice AI developers and researchers who want a flexible testbed for building and comparing voice assistants, and hobbyists building private, local voice interfaces.