Open-source, self-hosted AI chatbot for support, docs, and customer questions on any website.
Add one <script> tag, connect your content, and launch a branded assistant with RAG, guardrails, and multi-tenant control.
- Launch fast. Add one script tag, scrape your site or ingest files, and go from blank widget to working assistant quickly.
- Keep your stack and your data. Self-host it, bring your own model, and run on SQLite +
sqlite-vecwithout standing up a separate vector database. - Make it feel like your brand. Theme the widget, tune the welcome copy, set starter questions, and control handoff actions.
- Scale beyond one site. Run multiple branded assistants from one deployment with isolated content, prompts, and guardrails.
- Stay practical. Ship with built-in prompt-injection detection, topic controls, output filtering, and per-tenant rate limits.
Built with Hono, Vercel AI SDK, sqlite-vec, and Turborepo.
- Support teams that want fast answers on pricing, onboarding, returns, docs, or account questions.
- Product and marketing teams that want a branded AI assistant embedded on landing pages, docs, or help centers.
- Agencies and multi-brand operators that need one deployment serving many clients, regions, or business units.
Want a working local assistant fast? Requirements: Node.js 20+, pnpm 9+ (corepack enable pnpm).
git clone https://github.com/Maxkrvo/Chattr.git
cd Chattr
pnpm install
pnpm onboardpnpm onboard is interactive. It asks for your model provider (OpenAI, Anthropic, Azure OpenAI, or Ollama), API key, brand settings, and a site URL to scrape. It writes apps/server/.env and rewrites apps/server/src/instance/default/chattr.config.ts, then offers to seed bundled demo content so the bot can answer questions right away. Re-run it any time to update your setup.
Want full control instead of the wizard? Copy apps/server/.env.example to apps/server/.env, edit apps/server/src/instance/default/chattr.config.ts, then run pnpm seed-demo to load bundled FAQ content (or pnpm --filter @chattr/server scrape-ingest --tenant default to scrape your own site).
Build and run:
pnpm build
pnpm startOpen http://localhost:3000/demo.html to see the widget locally.
| Provider | Chat | Embeddings | Notes |
|---|---|---|---|
openai (default) |
yes | yes | Works out of the box. |
anthropic |
yes | via OpenAI | Claude for chat, OpenAI for embeddings. |
azure-openai |
yes | yes | Set resource name and deployment names. |
ollama |
yes | yes | Any OpenAI-compatible local server. |
Chattr also runs fully local against Ollama (or LM Studio, llama.cpp, vLLM). No keys, no bills, no data leaves your machine.
# 1. Install Ollama, then pull a chat and an embedding model:
ollama pull llama3.2
ollama pull nomic-embed-text
# 2. Install Chattr and pick `ollama` when onboard asks for a provider:
pnpm install
pnpm onboard
# 3. Build and run:
pnpm build && pnpm startOpen http://localhost:3000/demo.html. The bundled demo content is already embedded against your local Ollama so the bot answers immediately.
Switching embedding models later requires re-ingesting. Clear ./data/ first, then update OLLAMA_EMBEDDING_MODEL and OLLAMA_EMBEDDING_DIMENSIONS (see the deployment guide).
For most sites, embedding Chattr is a one-script change:
<script
src="https://your-server.com/widget.js"
data-server="https://your-server.com"
defer
></script>Useful attributes:
| Attribute | Purpose |
|---|---|
data-server |
Required. Chattr server URL. |
data-tenant |
Optional tenant ID (defaults to the server's default tenant). |
data-theme-primary |
Accent color override. |
data-theme-title |
Header title override. |
data-theme-subtitle |
Subtitle override. |
data-theme-avatar |
Avatar image URL override. |
data-language |
Force en or nl. |
If you want the widget to use tenant defaults (avatar, starter questions, escalation links), keep data-tenant aligned with the tenant keys in apps/server/src/instance/default/chattr.config.ts.
Feed Chattr from local files or let it crawl your site via sitemap-driven scraping.
Manual ingest:
pnpm --filter @chattr/server ingest ./path/to/docs
pnpm --filter @chattr/server ingest --tenant default ./path/to/docsScrape and ingest the default tenant:
pnpm --filter @chattr/server scrape-ingest --tenant defaultPreview scraped pages without writing to the vector store:
pnpm --filter @chattr/server scrape-ingest --tenant default --dry-runChattr is self-hosted. When you're ready to ship, Docker is the fastest path; Ubuntu VPS and reverse-proxy instructions live in docs/deployment.md.
The image auto-seeds bundled demo content on first boot when a supported API key is present, so a fresh container responds to questions immediately.
docker build -t chattr .
docker run -p 3000:3000 \
-e OPENAI_API_KEY=sk-... \
-v chattr-data:/app/data \
chattrOpen http://localhost:3000/demo.html to chat with the bot. Set CHATTR_AUTOSEED=0 to skip the first-boot seed if you're restoring a database into the chattr-data volume.
cp apps/server/.env apps/server/.env.local
pnpm devUseful scripts:
pnpm buildpnpm typecheckpnpm --filter @chattr/server check-retrieval -- --tenant default --query "How do I contact support?" --expect contactpnpm --filter @chattr/server report-chat-ops
Contributions welcome. See CONTRIBUTING.md for local setup and PR guidelines.
Security issues are handled privately via GitHub Security Advisories. See SECURITY.md for reporting details and scope.