Bolnee Chat

Any OpenAI-compatible API works. The README names OpenRouter, OpenAI, Groq, Ollama, and vLLM as examples. Provider, base URL, model, and API key are configured per bot in the dashboard, and a Fetch models button lists the models available for the chosen endpoint, prioritizing free models on OpenRouter. A global fallback provider can also be set via LLM_BASE_URL, LLM_API_KEY, and LLM_MODEL environment variables.

Each bot can ingest a website URL, uploaded files, or both. Supported file types are PDF, TXT, MD, DOCX, and FAQ content. The same-origin crawler respects robots.txt and extracts headings, paragraphs, and list items, and every source shows a processing status from queued through crawling, parsing, and indexing until it is fully indexed into SQLite full-text search.

You paste two lines before the closing body tag of any site: a script block defining window.BotConfig with your bot name, avatar, chat URL, accent color, greeting, and theme, plus a script tag pointing at chatbot-widget.js. The embed URLs are generated from your deployment origin automatically, and the widget renders as a floating bubble with a sliding chat window.

Yes. The Chats tab groups conversations by visitor and then by date, and offers downloads as CSV for Excel, JSON, or PDF via the browser print dialog. The dashboard also shows the number of active sessions, defined as distinct visitors in the last five minutes.

It runs locally or on your own server on macOS, Linux, and Windows via WSL, requiring Node.js 18+ and optionally Python 3.10+ for the crawler. The repository ships config files and instructions for deploying to Vercel and Cloudflare Pages or Workers. Data lives in a local SQLite file at data/bolnee.db, so no external database is required.