BrowserAI is an open-source library that runs production-ready LLMs and audio models directly inside the browser. It taps the user’s local graphics hardware through WebGPU to deliver fast, native-grade inference without touching a remote cloud server.
Instead of forcing you to wire up messy runtimes yourself, it provides a clean SDK that swaps between engines like MLC, Transformers, Flare, and Demucs under the hood. Heavy computation stays off the main thread thanks to built-in Web Worker support, keeping your frontend responsive while models generate text or process speech.
Running inference on the client shifts compute costs entirely away from your backend. Your API bills drop to zero because users bring their own hardware, which lets you scale apps without worrying about runaway infrastructure expenses.
It also solves data compliance out of the box: chat transcripts, documents, and voice recordings never leave the local machine.
Once the browser caches the model weights on the first visit, the app keeps working completely offline with zero network latency and no risk of third-party API outages.
Features
- 100% Local & Private: Zero data leaves your browser. No servers, no leaks.
- WebGPU Powered: Near-native speed with hardware-accelerated inference.
- Offline First: Download once, run forever without an internet connection.
- Multi-Engine Core: Seamlessly switch between MLC, Transformers, Flare, and Demucs.
- Audio Mastery: Built-in speech recognition, TTS, and AI stem separation (vocals/drums/bass).
- Structured Output: Generate clean JSON schemas directly from LLMs.
- Zero Infrastructure: Eliminate server costs and complex backend maintenance.
- Dev-Ready SDK: Simple API with Web Worker support for non-blocking UI performance.
