AI chat — think out loud on the board
A full conversation widget you can add more than once, each with its own provider and model.
A conversation widget that lives on the board, for when you need to think out loud without leaving what you are doing — drafting a reply, untangling a decision, rubber-ducking a problem at 11pm.
Using it
Type in the field, press Enter. The chat scrolls, the widget keeps its own history, and New chat starts a clean thread whenever the current one has drifted.
Connect your own key
Tap the model badge at the top of the card. Pick a provider — Anthropic, OpenAI, Google, xAI or Moonshot AI — paste your own API key from that provider, and Divergify fetches the live list of models your key can use. Pick one and you're chatting for real.
The key is yours: it's encrypted and stored so Divergify can proxy your requests, never shown back to you or to anyone else, and the spend is on your provider account — Divergify doesn't see a cent of it. Disconnect any time from the same menu; your chat history stays.
Add more than one
This widget is multi-instance: put several on the board, each with its own provider, key and model, and rename them so you know which is which — "Writing", "Code", "Ideas". Add another with + Add widget → AI Assistant. See The board.
AI chat vs Ask Divergify
They are different things and it is worth knowing which one you want:
- AI chat (this widget) is a general conversation, on a provider and key you connect yourself. It does not read your tasks, your calendar or your email.
- [Ask Divergify](/help/assistant) is the assistant in the top bar. It reads your actual board and can propose changes to it — always asking before it touches anything — and runs on Divergify's own account, at no cost to you.
"Write me a polite way to say no" is this widget. "What's left today?" is Ask Divergify.
When replies are simulated
Until a card has a provider connected, it answers with a visible demo reply that says so plainly. It is not broken and it is not pretending — connect your own key and genuine answers appear in the same interface, with nothing else to change.
See also
Updated 2026-08-11