What your AI saves to Cranes

Cranes learns from your own words in other AIs, never from the work itself or your AI’s replies.

Saved when you say so

Ask. “Save this to Cranes”, “remember that I’ve changed my mind about this”, “that’s wrong, I don’t think that at all”. Your AI saves your words as an entry and tells you in a few words where it went.
Say yes to an offer. After answering with your thinking, your AI may offer to save what you said, or Save to Cranes appears in the view of what Cranes guided. If you corrected your AI earlier in the conversation, the offer can include those words. When you agree, your AI adds one sentence about what you were responding to, so “no, the first one’s better” still makes sense later. That sentence is marked as the AI’s.
A reply you want kept. Ask for it, like “save that checklist”. It’s marked Written by that AI, never as your words. Agreeing with a reply doesn’t save it.
A skill. Ask your AI to save instructions you wrote, like an AGENTS.md, CLAUDE.md, or SKILL.md, as a skill. It’s kept word for word under its name. A newer skill with the same name is kept beside the older one and read as current. Cranes never writes a skill for you.

Kept quietly, if you allow it

With Keep my own ideas on, when you ask Cranes for your thinking and say something of your own, Cranes keeps your words to learn from, since you may never think to save them:

a position you take, or one you change;
a distinction or connection you draw;
a judgment with its reasons;
a correction of how Cranes or your AI read you;
a standard tucked into a request, like “make it plainer, I hate throat-clearing”.

Your AI doesn’t announce it, but tells you if you ask. Only your message is kept, never the conversation.

Never kept

Your AI’s replies, summaries, or plans as your words.
The work itself: code, changes, files, documents, logs.
Bare requests, like “run the tests”, routine logistics, messages that are mostly pasted material, and messages mainly about other people.
Passwords, keys, and personal details your thinking doesn’t need.
Health, money, relationships, and legal matters, and confidential details about clients, deals, or colleagues.
Someone else’s words as yours, like a teammate’s review comment.
Anything from a general question Cranes has nothing to add to.

Where it goes

What’s kept goes where the conversation is for: the team, a group, or, when you didn’t say, your own thinking, Just me. You can share it wider later. A correction of something Cranes brought in goes to the thinking it came from. See When Cranes comes in.

Your AI says something was kept, not learned. Your own thinking starts learning from it right away; the team’s or a group’s learns from it in its next pass, or right away when you asked to save.

Your entry settings hold here too, including Don’t learn from this and Keep off connected AIs. See Entry settings.

See, remove, or turn it off

Everything kept appears in Entries, marked with the app it came from. Open one and choose Remove now to delete its words.

To stop keeping your words quietly, in Use in other AIs (Settings › Connected AIs):

For one AI: open its row under Your AIs and turn off Keep my own ideas.
For every AI: under Your ideas, turn off Let connected AIs keep my ideas.

You can still ask your AI to save things. What was already kept stays until you remove it.