Every conversation follows the same path — and you can see where it went.
The scary part of an AI in your inbox is losing the thread. Donna is built so you can always see what it did and why — and take it back mid-conversation.
Donna runs on your own OpenAI or Anthropic key, so model costs are billed to you at cost — we never mark them up or resell tokens. Most support answers cost a fraction of a cent, and the bill is visible in your provider's dashboard, not ours.
No key yet? Paste ours for the first month and switch when you're ready.
You see it, the same as any other message in the thread. Every answer is logged with the sources it used, so you can correct the source and it won't repeat the mistake. Start in draft mode if you'd rather approve each reply before it sends — plenty of teams stay there for weeks.
No. Donna is an app you authorise; it writes into your existing conversations like a teammate would. Your widget, routing rules, triggers and team permissions are untouched, and removing the app leaves your inbox exactly as it was.
By your rules: topics you mark human-only, detected frustration, or a customer simply asking for a person. When it hands off, the thread is assigned with a one-line summary of what was already said.
Conversation text goes to the model provider whose key you configured, and nowhere else. We store transcripts so you can review them, we don't train on your data, and you can delete everything from settings at any time. EU data residency is available on Pro.
Your website is usually enough to start, and your last few hundred Crisp conversations teach it the rest — with your approval before anything is used. Most teams write two or three short notes to fill the gaps and stop there.
No. We're an independent company, and Donna is a third-party app distributed through official app marketplaces like Crisp, Zendesk and Intercom. Each name is a trademark of its respective owner and we're not affiliated with or endorsed by any of them.