Open source · v0.1

Drop-in AI chat for any website.

One <script> tag. Browser-side RAG. BYOK LLM. Source-citable answers. The AI chat widget that doesn't charge you per query.

View on GitHub See it in action

Why another chat widget?

10 KB widget

Tiny bundle. Transformers.js + the embedding model load lazily from a CDN only when a visitor opens the chat — never on page load.

No per-query SaaS bill

Inkeep and Kapa.ai charge $50–$500/mo for what amounts to an LLM proxy. You pay only for LLM tokens, only when someone chats.

BYOK LLM

The widget calls YOUR backend, never an LLM provider directly. Your key, your rate limits, your prompts. OpenRouter, Anthropic, OpenAI — all fine.

Browser-side embedding

Query embedding (all-MiniLM-L6-v2, q8) runs on the visitor's device. No embedding-API bill, no vector DB to host, no PII leaving the browser.

Source-cited answers

Every reply links back to the pages it drew from. Visitors verify with one click.

Themed, accessible

Light / dark / auto, primary colour, position, custom strings — all via data attributes. Namespaced CSS, no iframe.

Install in 3 steps

From zero to a working AI chat on your site in a few minutes.

Build your content index

Write a siteask.config.json pointing at your markdown directory or URLs, then run the CLI to generate a static siteask-index.json.

siteask.config.json

{
  "out": "./public/siteask-index.json",
  "sources": [
    { "kind": "markdown", "dir": "./content/blog", "urlPrefix": "/blog" },
    { "kind": "urls", "urls": ["https://your-site.com/about"] }
  ]
}

Deploy the tiny backend

A reference implementation for Next.js, Vercel Edge, Cloudflare Workers, and Express ships in the repo. Set OPENROUTER_API_KEY and you're done.

contract

POST  /api/siteask
{ query, contexts: [...] }

→ text/plain (streamed)

Drop the script tag

One script tag with data-index-url and data-api-url — the widget auto-mounts on page load.

any page

<script
  src="https://cdn.jsdelivr.net/npm/@siteask/widget@latest/dist/widget.js"
  data-index-url="/siteask-index.json"
  data-api-url="/api/siteask"
  data-title="Ask the docs"
  data-suggestions="What is X?|How do I install Y?|Pricing"
  async
></script>

Bring the backend you already use

Reference implementations for every common stack. ~100 LOC each — copy, set a key, ship.

Next.js API Route

Node.js

Vercel Edge Function

Edge

Cloudflare Worker

Workers

Express

Node.js

See it in action — on this very site

The floating chat button in the bottom-right of this page is the SiteAsk pattern hand-implemented for the portfolio. Same architecture: browser-side embedding via Transformers.js, retrieval over a pre-built index, an LLM call against a rate-limited backend route. Click it, ask anything about Amit's work, watch the source-cited answer stream in.

The dogfooded backend the public widget would call lives at /api/siteask and follows the exact public contract.

Roadmap

✓Widget bundle (IIFE + ESM) with auto-init from script attributes and SiteAsk.init() programmatic API.
✓Node CLI to generate siteask-index.json from markdown directories or URLs.
✓Reference backends for Next.js (Node), Next.js (Edge), Cloudflare Workers, and Express.
○npm publish to @siteask/widget so the install becomes a single script tag instead of clone-and-build.
○Optional WebGPU acceleration (~5× faster embedding on supported devices).
○Streaming retrieval — rerank as the visitor types.
○Conversation memory across turns.
○Adapter for Cloudflare Vectorize / Pinecone for site owners who want server-side indices.

Read the source on GitHub