Approve your first policy
The policy is the rulebook: which AI may write what, how much, and what it may forget. A new server starts with memory switched off until you approve one. You'll get a sensible draft to start from.
What the draft says, in plain words
Open the Policy tab. The draft is a text file with a comment above each part. Here's what it sets up:
| Who | May write | Per day | May forget |
|---|---|---|---|
| The brain Claude in your chats | Anything, and may pin notes to its core | 300 notes | Its own notes |
| Code Claude Code | Decisions, things it shipped and gotchas, only under your project folders | 60 notes | Nothing |
| You the admin | Only "forget" markers, from the admin page | No limit | Anything |
Some subjects belong to the brain alone, whatever else the rules say: self, people, ways-of-working and identity. Claude Code can read them but never change them.
The rest sets how quickly unused notes fade (half strength after 180 days, or 30 days for breadcrumbs, the "where we left off" notes) and how much the brain reads at the start of each chat.
The one line to change
Under code:, the draft lets Claude Code write to placeholder projects:
may_write_subjects: ["project-a/**", "project-b/**", "aingram/**"]
Swap those for the projects you use Claude Code on. Short, lowercase names with dashes work best. /** means "this folder and everything in it".
may_write_subjects: ["garden-app/**", "website/**"]
Not using Claude Code? Leave it as it is. It does nothing until you create a key for Code.
Approve it
- Click Check changes
The page checks the policy is valid and shows what you changed, in green and red. A mistake is caught here, before anything is signed.
- Click Approve in wallet
Your wallet shows a message starting
AIngram admin command. For a policy, it shows a fingerprint of the text rather than the whole text. The page has already checked that the fingerprint matches what you just reviewed. - Sign
A green banner confirms the policy is in force. Memory is now on.
Come back to the Policy tab, edit, check and approve again. Every approved version is kept, so you can always see what the rules were and when they changed.
Want a hand?
Your AI can explain any part of the policy or suggest a change. Copy the policy text from the page into a chat with this:
This is the policy for my AIngram, the long-term memory you use. Explain in plain words what it lets each voice do. Then suggest any changes for how I work: [say how you'd like it to work]. Show the changed lines exactly, keep everything else as it is, and say in one line why for each change.