Polymath.

Embeddable · grounded · capped

Your site can answer for itself.

Polymath is a chat widget that answers visitors from content you provide — installed with one script tag, capped to a budget you set.

index.html — the tag this page is running right now 
<script
  src="https://app.polymath-ai.live/polymath-embed.js"
  data-site-key="pk_poly_Gqr4OZVZCN7kgKdvPQJyXK"
  data-api-base="https://api.polymath-ai.live"
  data-turnstile-sitekey="0x4AAAAAAD0a6UQmZ8L0SZmZ"
  data-theme="auto"
  data-position="right"
  data-draggable="true"
  defer
></script>

That's the entire install, and it's live here — the chat on this page is this exact tag. Ask it anything; that's the demo.

Try it
Dress it

Answering in production on

From zero to answering

Three steps, in order. The last one is the script tag above.

01

Create a site

Each site gets a publishable key that identifies it — along with its own budget, rate limit, and theme.

pk_poly_…
02

Upload its knowledge

Docs, FAQs, product notes. Answers are retrieved from this material and nothing else.

docs → per-site datastore
03

Paste the tag

One script tag on any page — no SDK, no build step. The widget themes itself to your site.

<script … defer>

Built to face the public internet.

Anonymous visitors chat on your dime, so every message passes these guardrails — and when one can't be verified, the service refuses the request rather than failing open. Don't take the page's word for any of this: ask the widget.

Budget

A hard monthly spend cap, checked before every model call. At the cap, chat stops answering — your bill stops growing.

Throttle

Per-site and per-IP rate limits absorb bursts and bots long before they can become spend.

Challenge

Browsers chat on expiring tokens minted behind a bot challenge — the publishable key alone can't spend anything.

Grounding

Each site answers from its own uploaded material — not open-web model chatter in your brand's voice.

Isolation

Sites are isolated end to end, and secret keys are stored only as hashes — shown once, never again.

Privacy

Visitor identity isn't kept: chat runs anonymously and raw IP addresses are never stored.

Pay for answers, not pageviews.

One project subscription, one credit pool shared across its sites. Every plan gets the same provider-neutral routing and answer-quality target.

Planned launch pricing. The catalog is set; self-serve billing is not open yet.

$0.001 retail value
of one credit
1 credit per 1,000
input tokens
3 credits per 1,000
output tokens
+5 only when managed
retrieval runs
ordinary answer ≈ 5 credits grounded answer ≈ 10 credits

Free

Try a real site

$0/month

250 credits · preview first, then go live

  • 1 site · 10 docs / 1 MiB
  • About 25–50 preview or live answers
  • Basic analytics and lead capture
  • Powered by Polymath attribution
  • +500 live-test credits after verified install

Starter

Practical default

$9/month

5,000 credits · about 500–1,000 answers

  • 1 site · 200 docs / 20 MiB
  • Summary analytics and lead capture
  • Optional attribution
  • Paid credit top-ups

Growth

Small fleet

$29/month

20,000 credits · about 2,000–4,000 answers

  • 3 sites · 200 docs / 20 MiB each
  • One shared project balance
  • 90-day analytics
  • Higher lead and rate ceilings

Scale

Larger fleet

$79/month

60,000 credits · about 6,000–12,000 answers

  • 10 sites · 200 docs / 20 MiB each
  • One shared project balance
  • Usage export
  • Priority support

Actual use follows tokens. Answer ranges run from grounded to ordinary answers. Included credits expire monthly. Paid top-ups last 12 months: $10 / 10,000 credits, $20 / 22,000, or $50 / 57,500. No traffic meter, provider surcharge, or unlimited plan.

Pricing is set. Checkout isn't.

Polymath runs my own apps first. Self-serve billing is the next commercial layer — if you want it answering on your site before then, say hello.

Email Varun