An FAQ chatbot for your website that answers from your own documents

$ cat features.md

• Reads PDF, TXT, Markdown and HTML files or pasted text, splits them into passages (questions stay with their answers) and keeps them in your browser (IndexedDB).
• Hybrid search: keywords plus an optional small embedding model that runs in your browser, so a question worded differently from your FAQ still finds the answer.
• Answers with numbered citations; click one to see the exact passage, file and page it came from.
• Your own key for Anthropic Claude (Haiku 5.5 or Sonnet 5.5) or any OpenAI-compatible API, or a local model in LM Studio or Ollama. Requests go straight from your browser to that provider.
• No key yet? It still shows the matching passages, without an AI answer.
• Export a widget bundle for your site: one script tag, your index as JSON, a search-only mode that needs no server, and a small PHP or Cloudflare Worker proxy so your key never reaches visitors.

$ cat why.md

I built the retrieval chatbot for Einfach Fahrschule, a SaaS for driving schools, where it answers students from each school's own documents. Most small businesses need the same thing for their FAQ, without a monthly subscription or uploading their documents to yet another service. This is a fresh, small version of that idea that runs in the browser: your documents stay with you, every answer shows its source, and the widget you export is yours to host.

$ self-host-it

Requirements

Static files, no build step: any web server works (ES modules, so not from file://). The exported widget needs only two static files; AI answers need PHP 7.4+ or a Cloudflare Worker for the proxy.

Built with

HTML, JavaScript, transformers.js (in-browser embeddings); optional PHP or Cloudflare Worker proxy

License

MIT

The repository goes public on GitHub soon. Until then, ask me for the code and I'll send you access.