Bolnee-Chat: Self-Hosted RAG Chatbot Platform Works With Any OpenAI-Compatible Provider

A new open-source project appeared on Hacker News as a Show HN post: Bolnee-Chat, a self-hosted platform for adding retrieval-augmented chatbots to business websites. According to the project's GitHub repository, the tool is free to run, stores everything in SQLite, and works with any provider that exposes an OpenAI-compatible API.

What it does

The intended workflow has four steps. First, you create a bot and give it a name, avatar and greeting. Second, you add knowledge by pointing the platform at a website URL, uploading files, or both — supported formats include PDF, TXT, Markdown and DOCX. Third, you configure an LLM provider from a list that includes OpenRouter, OpenAI, Groq, Ollama and vLLM. Finally, you paste a two-line embed snippet — a small configuration object plus a script tag — before the closing body tag of your site.

The README positions the project against per-message billing and vendor lock-in: everything runs on your own infrastructure, and the only recurring cost is whatever you pay your chosen model provider.

How answers stay grounded

Bolnee-Chat is built around retrieval-augmented generation. A Python crawler walks your site within the same origin, respects robots.txt, extracts headings and body text, deduplicates content and writes the results to a JSON file. Crawled pages and uploaded documents are then chunked into SQLite's full-text search index. When a visitor asks a question, the system retrieves matching chunks, builds a grounded prompt, and streams the answer to the widget over server-sent events with the sources cited. A configurable fallback message covers questions where nothing in the knowledge base matches.

Dashboard, widget and chat history

The admin console covers appearance editing with live preview, per-bot statistics, knowledge management with a status pipeline that runs from queued through crawling and parsing to indexed, and settings for provider details, default messages and bot deletion. Visitor conversations are grouped by visitor and date, and can be exported as CSV, JSON or PDF.

On the public side, the embeddable widget appears as a floating bubble that opens a sliding chat window with light, dark or automatic theming. It keeps a visitor identifier and chat history in localStorage so the greeting shows once and conversations survive the window being closed.

Requirements and security notes

Running it yourself requires Node.js 18 or newer and npm; Python 3.10 with packages such as aiohttp, beautifulsoup4 and lxml is optional but recommended for the crawler, with Playwright available for JavaScript-heavy sites. The project can be deployed on a plain server, Vercel or Cloudflare Pages, and the SQLite database is created on first run.

Two implementation details stand out. Provider API keys are encrypted at rest with AES-256-GCM and, per the README, never appear in the embed snippet. At the same time, the quick-start configuration suggests setting an environment flag that disables authentication for the simplest single-user setup — anyone deploying the dashboard publicly should think carefully about that before exposing it to the internet.

Why it matters

Hosted chatbot services typically charge per conversation, per seat or per message, and they hold your visitor data. Bolnee-Chat inverts that model: the software is free, the data stays in a SQLite file on your infrastructure, and you choose the model — including fully local options like Ollama or vLLM for privacy-sensitive deployments. For developers who want a support widget grounded in their own documentation without wiring up a RAG pipeline from scratch, this looks like a practical starting point, with the usual caveat that a young open-source project deserves scrutiny before it handles production traffic.