Every grounded answer links to the page it came from. When the agent is not sure, it says so instead of guessing.
Every answer carries clickable citations, and a grounded mode that says I don't know instead of guessing.
Every model's cost in credits is printed on its card before you pick it, never discovered on the invoice.
Multi-agent and white label without a sales call. One workspace, one agent per client, your brand on the widget.
Websites, sitemaps, PDFs, docs, plain text and Q&A pairs. Midnight crawls, chunks and indexes it for your agents only, and re-syncs on a schedule so answers never go stale.
Every model card shows its credit cost per message up front. Run the same question on both tiers side-by-side in the playground and decide with evidence, not vibes.
Paste one script tag and the launcher is live on your site. The same agent answers on a hosted share link, in Slack, and on WhatsApp, with allowed domains keeping your key locked to your own properties.
Every "I don't know" becomes a to-do. Review the questions your agent missed, write the answer once as a Q&A pair, and it's covered for every visitor after that, no retraining ceremony.
Connect your systems and the agent acts mid-conversation, looks up an order, checks a subscription, books a call, opens a ticket. Every action is logged, scoped, and confirmed before it runs.
The channels it answers on, and the actions it can take. Anything without a prebuilt action is reachable as a custom HTTP call to your own endpoint.
Not used to train models. Not shared across workspaces. Deletable any time.
The differences that show up on your bill and in your answers.
The widget ships localized in seven languages, with full right-to-left layout for Arabic. The agent itself answers in whatever language the visitor writes in.
Credits are the only meter. Model multipliers are printed on the card.
Your first agent is free. Your first cited answer is about a minute away.