- tags
- #help #security #tokens #agents
- description
- AI tokens in AI Putty give your AI agents and tools a key to act as you on a project. Make one, see its agents, remint, suspend or revoke it.
- brief
- AI tokens is where you make a key for your AI or a tool, see the agents each role token minted, and remint, suspend or revoke it. A new value shows once.
API tokens for your AI agents¶✎ edit
AI tokens (Settings → AI permissions → AI tokens) is where you give your AI a key to act as you on this project, see who is using it, and take it back.
Your tokens¶✎ edit
There are two kinds:
- A role token (🤖) is for an AI. It mints an agent token of its own for each run.
- A tool token (🔧) is for a program, such as a sync plugin or a connector. It works directly, under its own name.
Role tokens come first, then Tool tokens. An admin also sees Everyone's (admin). Each is a table with Name, Project · role, Last used and Ends. Tap a heading to sort by it, again to reverse; type in the box under a heading to filter that column. Long lists show 20 a page with ‹ Prev and Next ›.
Find by name, role or note at the top searches all of them. A role token's row says how many agents are live; a tool token's says when it was last used. A red date ends within two weeks.
Make a token¶✎ edit
- Press + New token.
- Give it a Name (such as Claude on iPad) and pick its Kind: Role or Tool.
- Pick a Role. It sets what its writes are recorded as; it doesn't change what the token may do.
- Pick its Level and when it Expires: in 30 days, in 90 days, in a year, or never.
- Leave Its changes to topics are drafts on to review them on your drafts page (on the free plan it is always on). A role token can also be kept in your vault as
d2-<role>. - Press Create.
The value shows once. Press Copy, and copy the claude mcp add line too if you use Claude Code or another MCP client. Then press I've copied it. Missed it? Revoke the token and make another.
Open a token¶✎ edit
Tap a row to open it on the side:
- Token: its kind, project, role, level, when it was made, when it ends, last use, and a link to its secrets in the Vault.
- Agents minted from it (role tokens): live agents, each with Re-mint, Suspend and Revoke. Ended ones are folded by role, with Clear ended. Revoke its agents ends all live ones; the token stays. Re-mint agents gives them a fresh token with today's secrets.
- Remint: a new value for the token, keeping its secrets. Update lets the old value fetch the new one for a while; Rotate kills the old value now (use it if it may have leaked). Then tell your AI "collect your reminted d2 token" and put the new value where your AI keeps it.
- Name and note: rename it or say what it's for, then Save.
- Danger: Suspend pauses it and its agents until you Resume. Revoke ends it for good.
Revoke, Suspend and Clear ended ask once: press the Yes, … button to confirm.
Examples¶✎ edit
- "How do I connect a new AI chat to my project?" Press + New token, keep Role, pick a role such as designer, press Create, and copy the value to your AI.
- "I think a token leaked. What do I do?" Open it, choose Remint → Rotate, or press Revoke under Danger to end it at once.
- "Which of my agents are running right now?" Read the live count on each role token's row, then open it and tap Show under Agents minted from it.
- "My agent can't read a new secret." Open its role token and press Re-mint agents.
Good to know¶✎ edit
- A token never does more than you can, at most at the level you picked. See Security.
- Related: Vault, Permissions, Agent starts, Settings.