AI assistant trained on your content

How it works

Point the assistant at what you already have. Paste raw text, add a structured FAQ, drop in a list of URLs to crawl, or upload PDFs, Markdown, CSV and plain text files. Each source is chunked, embedded, and indexed into a per-project knowledge base — visitors never see a generic model answer, they see one grounded in your docs, pricing, and policies.

Replies stream token-by-token, the same way a human would type, so the widget never feels like it's stalled on a slow request. If the assistant isn't confident it has the right answer, it says so instead of guessing — you control the fallback message and the tone of voice through a custom system prompt.

Three ways to run it

• AI-only — the assistant handles every conversation end to end, ideal for FAQ-heavy support volume.
• Live agents — skip the AI entirely and route straight to your team inbox.
• Hybrid with smart handoff — the AI answers first and hands off to a human the moment a visitor asks something outside its confidence or explicitly asks for a person.

Bring your own model

Connect your own OpenAI API key and choose the model per project, so cost and quality are entirely in your hands — no shared rate limits, no black-box pricing markup.

Why it matters

Every AI answer is logged with the question, the retrieved context, and the reply, so you can audit quality over time, spot gaps in your knowledge base, and retrain by simply adding the missing content — no prompt engineering required for routine fixes.

Example

A SaaS company uploads its help-center articles and pricing page as sources. A visitor asks "do you support SSO on the starter plan?" — the assistant answers from the pricing page content, streams the reply live, and logs the exchange for the team to review later.