ventry089/llm-anywhere

llm-anywhere is a small backend that lets you add an AI chat feature to a static website without putting an API key in your frontend code, without signing up for a new AI provider, and without running or paying for a server. It's built as a single Cloudflare Worker, a small piece of code that runs on Cloudflare's edge network rather than on a machine you manage. The idea is simple: your website sends a message to the Worker, the Worker forwards it to a language model and gets a reply, then sends that reply back to your site. Because the Worker is the only thing talking to the AI provider, any secret key stays there and is never exposed in the browser where someone could copy it. By default, the project uses Cloudflare Workers AI to answer messages, which is free up to about 10,000 requests a day and needs no API key at all. Getting it running takes about five minutes: you clone the repository, log into a free Cloudflare account through the command line, and deploy the Worker, which immediately gives you a live web address you can send chat messages to. The README includes a simple example page you can open right after deploying to see a working chat interface. You can personalize the bot by editing two lines near the top of the worker.js file: one sets the system prompt that gives your assistant its personality, and the other picks which Cloudflare AI model answers the messages. If you want a stronger model than the free tier offers, you can add your own OpenRouter API key as a secret, and the Worker automatically switches to using that instead, with no code changes needed. The README is upfront about the limits. The free tier is capped at around 10,000 requests a day and uses mid-tier models, not the very best available. Cross-origin requests are wide open by default, so before putting this live you should lock it down to your own website's domain, or anyone could reuse your endpoint. It's a single proxied model call, not a full agent framework, and has no rate limiting against spamming, so add that yourself for a busier site. The project is released under the MIT license, which the README sums up simply as being free to do whatever you want with it.