Docs

Everything about setting up an assistant: what Docent is, how to train it, embed it, and what the plans include.

What Docent is

Docent is a chat widget you add to your website. It answers your visitors' questions using only the content you give it, and every answer shows the sources it came from. If the answer is not in your content, it says so instead of making something up.

Under the hood it reads your pages, splits them into small passages, and turns each passage into an embedding (a numeric fingerprint of its meaning). When someone asks a question, Docent finds the passages closest in meaning and asks a language model to answer using only those passages. That is why answers stay grounded in your content.

Adding your content

You can train an assistant on three kinds of source:

  • A URL. Paste any page address and Docent fetches and indexes its text.
  • A sitemap. Give it your sitemap.xml and it imports many pages at once (up to 40 per import).
  • A PDF. Upload a file (up to 15 MB) and Docent extracts the text.

Re-adding the same URL refreshes it, and a daily re-crawl keeps indexed pages current on their own, so answers do not go stale.

Embedding the widget

Add one line to your site, just before the closing body tag. The chat bubble appears automatically, themed to your brand.

<script src="https://docent.robinrahman.pro/widget.js" data-project="YOUR_PROJECT_ID" defer></script>

Your project id is shown on the project page in the dashboard. Each widget answers only from that one project, so different sites stay separate.

Customizing the look

In the dashboard under Appearance you can set the accent color, light or dark theme, header title and subtitle, greeting, starter questions, a logo, and whether the bubble sits on the left or right. A live preview updates as you edit, so the widget reads like part of your site rather than a bolt-on.

Choosing a model (bring your own key)

By default, answers use a shared model, which is the simplest way to start. You can also bring your own API key for OpenAI, Anthropic, or Google Gemini. When you do, calls are billed to you directly with your own rate limits, and your key is stored encrypted and used only on the server, never exposed to visitors.

Testing before you ship

Every project has a Test your assistant chat in the dashboard. It uses the exact same retrieval and answering as the live widget, so you can see precisely what your visitors will get before you embed anything.

Plans and limits

The Free plan includes 1 project, 100 questions per month, and 50 indexed pages, which is plenty to try it on a real site.

Pro raises that to 5 projects, 2,000 questions per month (unlimited when you use your own key), 1,000pages, and removes the “Powered by Docent” badge. It is $19 per month or $180 per year. See the pricing section for the full comparison.

Billing

Pro is paid in USDC (a stablecoin, so the price does not swing with the market). Because crypto cannot auto-charge a subscription, Pro is prepaid: you pay for 30 or 365 days and access simply expires at the end, so there is nothing to cancel. Paying again before it expires adds time onto what is left.

Privacy and security

Each project's content is isolated, so one assistant can never answer from another's pages. You can restrict which websites are allowed to use a widget with a domain allowlist, a per-day question cap protects against abuse, and any API keys you provide are encrypted at rest and used only server-side.