WebLLM: Run LLMs in Your Browser, No Server Needed
WebLLM is an open-source in-browser LLM inference engine from the MLC team, built on WebGPU. It lets users run large language models directly in the browser — no remote server, no API key, no internet connection required.
The project has over 18,000 GitHub stars and supports popular models like Llama 3, Mistral, Phi-3, and Gemma. Under the hood, it uses the TVM compiler for model optimization, achieving significantly faster inference than naive WebAssembly approaches.
For frontend developers, WebLLM provides a JavaScript SDK that integrates easily into web apps. Build an offline AI chat assistant, a local code completion tool, or a fully private document Q&A system. Model data is cached in the browser, so subsequent loads are fast after the initial download.
Highlight: Fully offline — data never leaves the browser. Supports multiple popular open-source models. Well-documented JS SDK with example code.
Kekoso: On-Device Mac Voice-to-Text
Kekoso is a macOS voice-to-text tool that processes everything locally — no cloud upload. It supports voice input in any Mac app, audio and video transcription, call recording, and YouTube link processing.
One-time purchase, no subscription, no account registration. Open the app and start using it.
For anyone who frequently takes voice memos, transcribes meeting recordings, or extracts podcast content, Kekoso offers a privacy-first local solution. Unlike cloud-based options like Google Docs voice input or Feishu Minutes, Kekoso keeps all data on your own device.
Highlight: On-device processing, data never leaves your machine. One-time purchase. Supports audio/video/YouTube multi-format transcription.
AQ: Multiplayer AI Coding Environment
AQ is a multiplayer agentic development environment built for AI coding agents. Developers can run multiple AI coding agents simultaneously in their own cloud, collaborating on code writing, debugging, and deployment.
Unlike single-user AI IDEs like Cursor or Windsurf, AQ's core concept is "multi-agent collaboration." Launch multiple coding agents in parallel, each handling different modules or tasks, sharing context and coordinating progress. Suited for team collaboration or parallel development of complex projects.
AQ supports self-hosting, keeping code and data on your own infrastructure. For teams concerned about data security and customization needs, this is a compelling choice.
Highlight: Multi AI-agent parallel collaboration. Self-hosting support. Designed for teams and complex projects.