Launching today
LocalMode
Run AI in your browser - no servers, no API keys
4 followers
Run AI in your browser - no servers, no API keys
4 followers
LocalMode is an open-source (MIT) toolkit for running AI entirely in the browser: LLM chat over 76 models, RAG, speech, and vision. No servers, no API keys, and data never leaves the device. WebGPU with a WebAssembly fallback; works offline.
Hey Product Hunt 👋
I built LocalMode because "AI in your app" almost always means shipping your users' data to someone else's servers, wiring up API keys, and paying per token.
I wanted the opposite: AI that runs entirely in the browser, where the data never leaves the device.
LocalMode is an open-source (MIT) toolkit for exactly that:
• LLM chat over 76 models across 5 runtimes (Transformers.js, WebLLM, wllama, LiteRT, MediaPipe) + Chrome Built-in AI
• RAG and vector search - embeddings, a vector DB, and semantic search
• Speech - Whisper speech-to-text and Kokoro text-to-speech (29 voices)
• Vision - image classification, object detection, and CLIP image search
• A zero-dependency core, 64 React hooks, and a shadcn UI registry of 107 components + 36 installable blocks
• Runs on WebGPU with a WebAssembly fallback, and works offline after the first model download
No servers. No API keys. No telemetry. It's MIT-licensed and the whole thing is
copy-owned - install a block with the shadcn CLI and you own the .tsx.
Try the live blocks (chat, RAG, vision, voice - all running in your tab): 👉 https://localmode.ai/blocks
UI Components: https://localmode.ai
Docs: https://localmode.dev
GitHub: https://github.com/LocalMode-AI/...
I'd love feedback on which capability you'd reach for first, and what's missing.
Happy to answer anything in the comments today 🙏
LocalMode looks great, been wanting something like this for a while. One thing that would help me a lot is a simple way to share or export my chat history and indexed docs, maybe as a single zip or encrypted bundle, so I can move setups between devices without redoing everything from scratch.
Running models fully in the browser is genuinely cool, and the offline angle is what sold me. One thing I'd love to see is a way to pin specific models as favorites with a simple localStorage flag so the chat interface remembers my preferred default across sessions. Right now I have to reselect the model every time I reopen the tab.
Tried the Gemma model in a private window and the WebGPU setup was surprisingly painless. Genuinely cool that speech and vision run locally without any keys or config.