Repolode
Submit
midudev

midudev/runonweb.ai

Browser-based open-source AI modules (speech, background removal, embeddings, translation, TTS) that run on-device via WebGPU or WASM, with an npm package.

View on GitHubWebsite
Stars
776
Forks
76
Open issues
0
Last commit
14 days ago

What is it?

runonweb.ai is a collection of open-source AI modules that run entirely in the browser, along with the `runonweb` npm package that exposes them. Each task has its own import path, such as speech-to-text, background removal, image captioning, depth estimation, object detection, text embeddings, classification, translation, emoji prediction and text-to-speech.

The modules use WebGPU when the browser supports it and fall back to WASM otherwise, and inference happens on the user's device. Each module is built on a permissively licensed open model, and the demo pages credit the base model, its author and its license.

The repository is a monorepo: the Astro site lives in `packages/web` and deploys to Cloudflare Workers, while the library lives in `packages/runonweb`. The project also documents backend quirks found during testing, such as which quantization formats fail on WebGPU and which models are pinned to WASM.

Key features

  • 01Separate imports for speech-to-text, background removal, captioning, depth, detection, embeddings, classification, translation, emoji and TTS.
  • 02Uses WebGPU when available and falls back to WASM.
  • 03Inference runs locally in the browser, so input data is not sent to a server.
  • 04Each module is built on an Apache-2.0 or MIT licensed open model.
  • 05Includes a helper for checking WebGPU availability and a cosine similarity function for embeddings.

Use cases

  • Adding in-browser speech transcription or text-to-speech to a web app without a server-side API.
  • Removing image backgrounds on the client for privacy-sensitive photos.
  • Running semantic search or similarity checks on text entirely in the browser.
  • Prototyping on-device AI features and comparing WebGPU and WASM performance.

Quick start

  1. 1

    Install dependencies with pnpm

    pnpm install
  2. 2

    Start the local dev server

    pnpm dev

Things to know

  • The project metadata lists no license, so reuse terms for the code are unclear.
  • Some modules run only on WASM, which is slower than WebGPU.
  • The caption model's LFM license is free only under USD 10M in revenue.
  • The classify large model is about 2.7 GB to download.

In collections

Run LLMs locallyAI video & audio tools

Topics

#webgpu#local-ai#web-ai#onnx#browser#embeddings#speech-to-text#wasm#ai-model#ai-models

FAQ

What is runonweb.ai?
Browser-based open-source AI modules (speech, background removal, embeddings, translation, TTS) that run on-device via WebGPU or WASM, with an npm package. runonweb.ai is a collection of open-source AI modules that run entirely in the browser, along with the `runonweb` npm package that exposes them. Each task has its own import path, such as speech-to-text, background removal, image captioning, depth estimation, object detection, text embeddings, classification, translation, emoji prediction and text-to-speech.
Is runonweb.ai free to use? What license does it have?
runonweb.ai has no declared license. Its code is public, but without a license you have no explicit permission to reuse it; ask the author before relying on it.
Is runonweb.ai still maintained?
Actively maintained. The last commit to runonweb.ai was on Oct 24, 2026.
What can I use runonweb.ai for?
Adding in-browser speech transcription or text-to-speech to a web app without a server-side API. Removing image backgrounds on the client for privacy-sensitive photos. Running semantic search or similarity checks on text entirely in the browser. Prototyping on-device AI features and comparing WebGPU and WASM performance.
How do I install runonweb.ai?
Install dependencies with pnpm: pnpm install Start the local dev server: pnpm dev
Where can I get runonweb.ai?
The source code and installation instructions are on GitHub: https://github.com/midudev/runonweb.ai

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

Related projects