Most Common Questions

No. The wizard builds the dataset and the assistant for you, and putting it on your site is a copy-paste snippet — or a one-click plugin on WordPress. If you can fill in a form and paste a block of text, you can ship an assistant.

The wizard itself takes a few minutes. The wait is the crawl: reading a small site is quick; a large one takes longer because it visits your pages one by one. You can keep working and come back when it finishes.

Yes. On Home, choose Ask your documents instead and upload files (a handbook, a price list, a set of PDFs). The assistant learns from those.

Yes. A memory dataset can mix crawled pages, uploaded files, and content added through the API (an Enterprise feature) — the assistant searches all of it together.

Only from your content — the dataset you connected. The assistant grounds each answer in what it finds there and can show the sources it used. It isn't answering from the open internet.

Almost always a content gap: the answer isn't in the dataset, or the page that has it wasn't crawled. Add or fix that content and, if needed, re-index the dataset — then ask again.

Yes. In the assistant's Behaviour settings, give it explicit rules like "Answer only from our content; if you're unsure, say so." Clear, short rules work better than long ones.

Feed it the right content, keep the behavior rules tight, and read Analytics to see what visitors actually ask — then close the gaps you find.

The assistant follows the language of your content and your behavior instructions, and the chat interface can be localized per assistant from the Translations settings.

Use Share: send an email invitation and they get a private link to chat with the assistant — no sign-up, no code on their side. See Share your assistant.

Two usual causes: your site's domain isn't in the assistant's allowed origins yet, or the security mode expects an access token it didn't get. Check both and reload; the browser console usually points at which one.

None for a public website, API (a key) for a private one, and OIDC if you already have an identity provider.

Only if the security mode is API (or OIDC). A None-mode assistant loads with just its ID — no key required.

Yes — there's a short guide for each. See Add your assistant to your website.

Yes — 15 days, no credit card required. Add a payment method whenever you want to continue; if you don't, the trial simply pauses when it ends.

Create one assistant, one data capture job, and one memory dataset, with website crawls at the Preview scope (up to 100 pages).

On usage — storage, LLM tokens, and data capture job proxy bandwidth. Basic and Pro bundle a monthly allowance; Enterprise is fully pay-as-you-go. See the pricing page.

Your subscription pauses automatically — nothing is charged by surprise. Add a payment method to pick up where you left off.

Put the data capture job on a schedule (or re-run it), and re-upload changed files. New content becomes answerable once processing finishes. You can also set a content lifespan so pages are re-processed automatically as they age.

Runs are capped by the crawl scope: Preview 100, Small 500, Medium 1500, or Large 5000 pages. Trial workspaces use Preview.

PDF, Word (DOCX), Excel (XLSX), PowerPoint (PPTX), text (TXT), rich text (RTF), Markdown (MD), and ZIP archives.

Re-index the dataset so the new content is searchable, and confirm the data capture job actually re-captured the changed page (check its content lifespan and schedule).