The questions you’re actually going to ask.

The bot won't write a sentence it can't cite. If your docs don't cover the question, it refuses with the closest related pages — visibly, in a different colour from a normal answer. If the bot does write an answer, every sentence links to the page it came from, so QA can spot drift fast and roll back. We've never had a confidently-wrong answer escape to a customer in pilot.

Every reply renders sentences with small clickable chips next to them. Click a chip, and the widget opens the exact page on your docs that the sentence came from. In your QA dashboard, the same applies — click any sentence in any conversation transcript and the source page opens. Citations are real URLs to real pages, not abstract references.

The bot flags this in your weekly content report: 'These two pages give different answers for X — choose one.' Until you resolve it, the bot will cite whichever page the search ranked higher (with the source visible to the customer). Most teams find this is the most useful side benefit of deploying — it surfaces drift in your documentation.

Yes. Per-source permission scopes let you mark sections as off-limits (e.g., internal HR policies, draft pricing, pre-announcement features). The bot won't index those, and asking about them triggers a polite refusal pointing the customer to the right human channel. Configurable per source on Scale and above.

Anything reachable as a public web page — your help center, public docs, blog, pricing page, product pages. We crawl JavaScript-rendered sites (most modern docs platforms) the same way a real browser does, so React/Next.js/Vue-rendered content comes through fully. We also support PDF, Markdown, and uploaded files on Scale and above.

Yes, on Scale and above. Point us at a sitemap with appropriate authentication (a service-account session, an authentication endpoint, or a per-page API key) and we'll crawl behind your login. Common pattern: a customer help center where logged-in customers see plan-specific content. The bot serves the right content based on each visitor's session.

The bot answers in 40+ languages, matching whatever language the customer asked in — even when your docs are English-only (it'll translate the answer and link to the source page). For docs with official translations, we crawl each locale separately and route by Accept-Language. Multi-locale docs are a Scale-tier feature.

The bot recrawls automatically — by default once a week, configurable as often as every 15 minutes on Scale and above. We only re-read pages that changed (the rest get skipped fast), so a typical recrawl runs in minutes and costs cents. Within an hour of publishing, the bot is answering with the updated content.

First crawl is under 60 minutes from sign-up. The widget snippet is two lines of HTML — your developer can paste it during a coffee break. Phased rollout to one channel or one ticket category typically ships in week 2. The 6-month rollouts you remember from custom AI projects don't apply here — we ship as a product, not a build.

Yes. The bot lives in front of your existing helpdesk: it handles tier-1, hands the rest to your human queue with full conversation context. No replacement of your existing CRM, no migration of your historical tickets, no impact on your macros and SLA timers. We have native integrations for the major helpdesks and a webhook fallback for everything else.

Shadow mode. Run the bot against your live ticket queue for 1–2 weeks without exposing it to customers — you see exactly which tickets it would have handled, with what answer, and what the projected CSAT looks like. Phased go-live to one channel or one ticket category. Most teams keep their full headcount and redirect saved hours to proactive work.

Your content is stored in a per-tenant encrypted prefix; cross-tenant access is a privilege violation and refused at the infrastructure level. EU and AU data residency available on Enterprise. We don't share or sell data, period. The Sub-processor list is in our DPA appendix — sent to procurement on request, with 30-day notice on any change.

No. Your content is retrieved at inference time, not learned. We don't fine-tune models on tenant data. This keeps you out of the 'high-risk AI system' classification under the EU AI Act for content-grounded customer service, and it means your competitors don't benefit from your docs being on our platform.

SOC 2 Type I report under NDA, Type II in audit observation. ISO 27001 in progress. Signed DPA with Standard Contractual Clauses available — DocuSign turnaround under one business day. EU AI Act Article 52 transparency requirements pre-cleared. The full security pack and methodology lives at /security.

Plans are priced on monthly answered questions, with overage at $0.005 per answer above plan. A 5x traffic spike turns into a $250 overage line, not a contract renegotiation. Annual contracts can include a TCO ceiling — your CFO is not getting surprised. The ROI calculator on /pricing shows the math against your real volume.

Starter ($49/mo, 1 source, 5K answers) is for small docs sites. Growth ($199/mo, 3 sources, 25K answers) is the typical SMB SaaS plan. Scale ($499/mo, 10 sources, 100K answers) adds multi-locale, per-source permissions, API access. Mid-market ($1,499/mo) adds higher rate limits and premium model routing. Enterprise (from $2,500/mo) adds SAML/OIDC, dedicated support, custom SLAs, contractual TCO ceiling.