Show HN: Bolnee-Chat – Self Hosted Chatbot Integration in Your Business Website

Chatbot integration in your business website. Self hosted, free forever!

Bolnee-Chat is a self-hosted RAG chatbot platform. Create a bot, add your website + PDFs as knowledge, pick any OpenAI-compatible provider (OpenRouter, OpenAI, Groq, Ollama, vLLM), and embed a 2-line snippet. Answers are grounded in your sources with citations, visitor chats are grouped and exportable, and everything runs on your infrastructure with SQLite.

Why Bolnee-Chat?

All Features

Bot Overview — Stats & Snippet Per-bot: Live status, message/user counts, creation date, source count, embed snippet with botName/avatar/chatUrl/accent/greeting/theme.
Appearance — Live Preview Edit bot name, avatar (upload/preview), accent/background colour, theme light/dark/auto, greeting. Live preview; saved via PATCH /api/chatbots/:id.
Chats — Grouped by Visitor Grouped by visitorId → IP, then by date, chronological. Refresh + download CSV (Excel) / JSON / PDF (print). Active sessions = distinct visitors last 5m.
Settings — Provider, Prompts, Danger Zone provider/model/baseUrl/apiKey, default message (pre-first-turn), fallback message (no sources matched), delete bot + chats + sources + chunks.
Widget — Floating, Theme-Aware, Persistent Bottom-right bubble → sliding window (360×520, 75vh mobile). Header with accent + avatar, typing dots, SSE streaming, sources cited, VISITOR_ID + history in localStorage so greeting shows once and chats survive close/open. Theme light/dark/auto (#0f172a / #fff).

Also in the box:

• Knowledge management — lists locator · type · status · error · date with delete; append more via Add knowledge.
• Avatar file storage — data URLs converted to /api/public/avatar/:id (2MB limit, data/avatars/).
• Encrypted provider keys — per-bot apiKey/baseUrl/model via AES-256-GCM, never exposed in snippet.
• Dark-mode console — #020617 bg, #1e293b cards, slate-800 inputs, no white surfaces.

Workflow

1. Create Chatbot — Brand in Seconds Name + avatar (PNG/JPG/WEBP ≤2MB, preview) → stored as /api/public/avatar/:id and shown in the widget header.
2. Add Knowledge — Website + Files URL-only, files-only, or both. Same-origin crawler (crawler/crawler.py) respects robots.txt, extracts h1/h2/p/li, dedups, saves data/{chatbotId}_website.json. Status: queued → crawling → parsing → indexing → indexed.
3. Configure Provider — Any OpenAI-Compatible API Pick provider → Base URL auto-fills → paste API key → Fetch models lists live models (prioritizes :free for OpenRouter). Keys stored encrypted (AES-256-GCM), never in embed.
4. Embed — 2 Lines Copy from Overview → Embed code or Knowledge → Step 4. Auto-configures origin via window.location.origin; widget loads via chatbot-widget.js with SSE streaming.

Requirements

Tech Stack

Installation

# Clone
git clone https://github.com/AniketWathore/bolnee-chat.git
cd bolnee-chat

# Env
cp .env.example .env
# edit .env — simplest self-hosted:
# DISABLE_AUTH=true
# JWT_SECRET=change-me-32-chars
# LLM_BASE_URL=https://openrouter.ai/api/v1   # optional global fallback
# LLM_API_KEY=sk-or-v1-...                     # optional global fallback
# LLM_MODEL=inclusionai/ling-3.0-flash-fin:free

# Install
npm install

# Python crawler deps (optional)
pip install aiohttp beautifulsoup4 lxml requests brotli

# Verify
npm run lint    # tsc --noEmit
npm run build   # vite + esbuild → dist/
npm run dev     # http://localhost:3000 (auto-fallback to 3001 if busy)

SQLite is created at data/bolnee.db (git-ignored). Crawled sites → data/{chatbotId}_website.json, chunks in SQLite.

Usage — Dashboard Flow

1. Create chatbot — + New chatbot → name + avatar (preview ≤2MB).
2. Add knowledge — enter https://your-site.com and/or upload PDF/TXT/MD/DOCX/FAQ → status queued → indexed.
3. Configure provider — choose provider → Fetch models → pick model → Save. Keys encrypted, not in embed.
4. Embed — copy snippet from Overview:

<script>
  window.BotConfig = {
    botName: "Customer Bot",
    avatar: "https://your-domain/api/public/avatar/BOT_ID",
    chatUrl: "https://your-domain/api/public/chat/BOT_ID",
    accentColor: "#111111",
    greeting: "Hi! How can I help?",
    theme: "dark"
  };
</script>
<script src="https://your-domain/chatbot-widget.js" async></script>

Paste before </body>. Widget stores VISITOR_ID + chat history in localStorage; greeting shows once.

Bot Console Reference

API Reference

Streaming: public/chatbot-widget.js:311 reads SSE via getReader(), falls back to text() + SW bypass for locked streams. Visitor grouping via X-Visitor-Id (VISITOR_ID in localStorage).

File Reference

Hosting

Dashboard is fully API-driven (/api/* relative) and auto-configures window.location.origin for embed URLs.

Vercel:

# env
DISABLE_AUTH=true
# JWT_SECRET not required for simple mode

# vercel.json already: rewrites /api/:path* → /api, /(.*) → /index.html, outputDirectory: dist
npm run build && vercel --prod

Cloudflare Pages / Workers:

# wrangler.toml: bucket = "./dist", DISABLE_AUTH=true
npm run build
wrangler pages deploy dist
# or: npx wrangler deploy

Chat endpoint streams SSE; for external sites use public https:// chatUrl (not localhost).

Configuration

.env.example documents all.

Troubleshooting

Verification Checklist

npm run lint      # tsc --noEmit clean
npm run build     # vite + esbuild → dist/
npm run dev       # http://localhost:3000
# Manual:
# 1. Create bot → avatar preview → Save appearance → preview updates
# 2. Add knowledge: URL + PDF → status queued → indexed
# 3. Provider → Fetch models → pick :free → Save
# 4. Overview → Copy embed → paste in plain HTML → widget loads, greeting once, close/open keeps history, dark/light/auto themes correct
# 5. Chats → grouped by visitor → CSV/JSON/PDF export
# 6. Settings → default/fallback messages → Chat without sources returns fallback

License

Distributed under the MIT License. See LICENSE for more information.