AskBox is a chat widget you can add to any website with a single <script> tag. It answers from your own docs, pages, and content — not from generic LLM hallucination — and falls back to email or a live handoff when the answer isn't there. No agent dashboard to staff, no monthly seat tax.
No SDK. No npm install. No tracking pixel that fights your CSP. It's one script tag and a domain.
<script src="https://yourcompany.askbox.io/widget.js"></script>AskBox crawls your site to build the knowledge base. You can also paste in URLs, upload PDFs, or wire up a sitemap feed for ongoing sync.
The major support widgets are great if you have a support team. AskBox is for when you don't — or for when you do, but the team is buried under repeat questions that have answers already documented somewhere.
| Intercom | Tidio / Crisp | AskBox | |
|---|---|---|---|
| Install | JS snippet + identity verification | JS snippet | One <script> tag, no identity setup |
| Answers from your own content | Yes — with Fin AI add-on, per resolution | Limited — separate AI add-on | Built in. Crawls your site or accepts uploads. |
| Pricing model | Per seat + per AI resolution | Per seat tiers | Flat per-site, no seat counting |
| Requires staffed inbox | Designed around live agents | Yes, for full value | No. Falls back to email when AI is uncertain. |
| "I don't know" behavior | Hands off to human | Hands off to human | Honestly says it doesn't know, then offers email or handoff. No invented answers. |
| Best fit | SaaS with a dedicated support org | E-commerce with live agents | Small teams, docs-heavy products, anywhere repeat Qs eat the inbox |
The chat bubble in the bottom-right of this page is AskBox, configured against the Coyote IT site itself. Ask it anything you'd want to know about our services, our products, or our pricing — including the stuff it shouldn't know, just to see how it handles questions outside its scope. (Spoiler: it says so.)
👉 Bottom-right corner. We'll wait.
Answers are grounded in retrieved chunks from your indexed content. If retrieval comes up empty, the model is told to say so — not to fill in the blank.
You set how confident the model has to be before it answers. Below the threshold, it routes to a configured fallback (email, calendar link, your existing support inbox).
Users see the doc / URL the answer came from. If the source is wrong, you fix the doc — not the chatbot's prompt.
Want help wiring AskBox up to your docs and site? Start with a free audit.
Book a free audit