AI chat

Set up AI chat once, then ask it questions or ask it to build a widget. AI chat is on the desktop app only.

Before you start: either a command-line AI tool installed on your computer, or an API key from Anthropic, OpenAI or Cursor. TeamHUD ships no model of its own.

Set it up

1. Open the chat. Hover the AI Chat hub on the overlay. A new install shows “AI chat is off. Nothing is sent to an AI model until you turn it on.”
2. Turn it on. Choose Turn on AI chat. It is off by default on purpose, and you can switch it off again in Preferences > AI Chat.
3. Open settings. Use the settings button in the panel header, or open Preferences > AI Chat. Both show the same settings.
4. Pick a provider. Under Provider, choose one that is not greyed out:
   • Claude (local CLI) is available when the claude command (Claude Code) is found on your computer. Sign in with Claude Code if you have not.
   • Cursor is available when the cursor-agent command is found. Run cursor-agent login if you have not signed in.
   • Claude (Anthropic) and OpenAI are available once you add a key.
5. Add a key if you chose one. Under API Keys, paste it into Anthropic, OpenAI or Cursor. The field then shows “✓ saved”. Paste a new key to replace it.
6. Choose a model (optional). Leave Model on Default to use the provider’s recommended model.

What you’ll see: the chosen provider selected, and each greyed-out provider says why (for example “No API key set” or “claude CLI not found — install Claude Code”).

Provider choices in Preferences > AI Chat.

Note

API keys stay on your computer and only go to the provider you chose. Settings below the switch add more control: Include teammates’ names and roles, Search the web without asking and Let AI read Slack when I ask. Each is off until you turn it on, and the last three also need your organization to allow them.

Ask your first question

1. Open AI Chat. Hover the hub. Or say “new AI chat” with Voice assistant.
2. Type your question in the message box and send it.
3. Read the answer. Your chats are saved as threads you can reopen. Delete a thread with its delete button, after confirming.

What you’ll see: a “thinking” indicator, then the answer. The hub shows a badge while the AI is working, when it needs your approval, when a turn failed, and when a finished answer is waiting.

How it differs from team chat

	Team chat	AI chat
You talk to	Your teammates	An AI model
Who sees it	The people in the conversation	You, and the provider you chose
Needs	A team, signed in	The switch on, plus a local tool or a key

Build a custom widget from chat

Ask for the widget in plain words, for example “make a widget that shows a countdown to Friday”. The AI proposes it, and nothing happens until you approve.

1. Ask. Describe the widget.
2. Review the card. A card reads Create widget “name”? and says the widget is sandboxed and what it can reach on the network, if anything.
3. Pick how much to edit. Choose Defaults, Basic (name and position), Medium (adds size) or Detailed (adds description and tags).
4. Approve. Choose Create, or Cancel. Changes and removals use Update and Delete the same way.

Next: Custom widgets.

If it goes wrong

• “Add an API key in settings to start chatting.” No provider is ready. Do steps 4 and 5 above.
• A provider is greyed out. Its tool is not installed or not on your PATH, or no key is saved. Install it, then reopen settings.
• The local tool answers with a sign-in error. Sign in to Claude Code or cursor-agent in a terminal, then ask again.
• The AI cannot use a TeamHUD tool. Providers marked tools can; check the chip next to the provider name.