InBrowser is a complete AI workspace that runs in a browser tab. Models, code execution, storage, agents and skills all execute on your own device. There is no account to create, no API key required to start, and no server of ours that sees your conversations.
Over 160 open models can be downloaded once and then run locally on WebGPU through WebLLM, with no network call and no rate limit afterwards. Chrome's built-in Gemini Nano works with no download at all. A local model has no key, no quota, and keeps working with the network switched off.
Several hosted providers work with no signup and no API key. Requests go directly from your browser to the provider rather than through a shared server, so rate limits apply per person instead of being exhausted by everyone at once. You can also point the app at your own OmniRoute gateway, or route traffic through an HTTP, HTTPS, SOCKS5 or SOCKS4 proxy you control.
Twelve language runtimes are compiled to WebAssembly and install into the browser on demand, including Python, JavaScript, C, C++, Ruby, PHP, R, Lua, Clojure and SQL. The assistant can execute code in any installed runtime and read back the real result, sandboxed with a timeout, instead of describing what the code would do.
Agents combine a system prompt, a model policy and a toolset into a bounded plan-act-observe loop that reports every step it takes. Skills are reusable instruction packages in the standard Agent Skills format, installable from any public marketplace repository. Tool plugins give the model exact utilities to call for regular expressions, diffing, JSON, CSV, hashing, JWT, colour, dates and units.
Conversations, skills, memories, scheduled tasks and settings are stored in your browser's own IndexedDB, localStorage, OPFS and Cache Storage. Provider API keys stay on your device and are deliberately excluded from backup exports. Everything can be exported as a single archive and imported again elsewhere.
Downloading InBrowser