Upgrade to a chat loop

Today you talk to your AI in its own app. In a chat loop it stays on d2 and answers you in the chat panel, beside the page you're looking at. One onboarding prompt connects it and starts the loop.

1
Let your AI reach d2Allow *.aiputty.com in its network settings, if it asks.
Claude

The Claude app seems written by a 5 year old with a degree in un-usability, and things move around on a daily basis… What seems to work at the present time: go to the web version, claude.ai, then

profile → Settings → Capabilities → Allow network egress → Domain allowlist

and either turn on All domains if you dare, or just add *.aiputty.com to the allowed list.

Memory, on the same Settings page:

profile → Settings → Memory → Generate memory from chats

Turn it on, and Search and reference chats too. On a Team or Enterprise plan an owner has to allow memory first. In the Claude app (desktop or phone) it should be the same path, from your name or profile picture → Settings → Memory; if it isn't there today, use the web version as above.

Also turn on your AI's memory, so its chats remember things between them (what you're building, how you like things done): it makes life much easier. How, for Claude: in the fold above.
2
Paste this into a chat with your AI
Claude on your desktop, web or phone. That chat becomes your loop.
```
Making your onboarding prompt…
```
💡 This works best in a Code section of the Claude app.
3
Connected
Waiting for your AI's loop…
When it's listening, say hello in the chat panel (bottom right, on any page) or open the pad: it answers in a few seconds.
Another chat, same AI?In any other chat where your AI already works on this project, paste this and it starts a loop there too:
```
start loop
```
A chat that has never connected to this project needs its own onboarding prompt: press New onboarding prompt in step 2. To stop a loop, type stop loop in its chat.
4
Handover

How the loop works

Where you talk: the chat panel on every d2 page, not your AI's app.
How it asks you things: TurboChat. Its options come as a pad: tap to pick, one CONTINUE sends your answer.
While it runs: keep that AI chat open (computer or phone); it uses your AI as usual, only while you chat.
Stop any time: type stop loop in that chat, or close it. When its chat gets full, it tells you; make a new onboarding prompt here for a new chat.

Mind Meld: let it show you

With Mind Meld on, your AI doesn't just tell you what it made: it takes this browser tab to the page it's talking about and outlines what changed, with Back to return to where you were. You keep talking in the chat panel; d2 follows.

Only your own AIs, and the agents they started, can move your tab. Nobody else's can.
Off by default, per tab: the Mind Meld button in the top bar turns it on or off any time, and so does Meld in the chat panel.

If it doesn't connect

No light after a minute: check your AI read the onboarding prompt. An onboarding prompt works once, for an hour: press New onboarding prompt and paste again.
Your AI says it can't reach the site: step 1.
Rather type in your AI's own app? Keep using your prompt.

Ready?

Start playing: your dashboard becomes this project's home.