Skip to product

chained.tools

agentpass

A vault so your agent can act without the secret entering the model. 1Password assumes a human is looking. Agents look by default.

Fail closed

Grant the action.

Keep the value in the vault.

  1. agentdeploy to prod, needs GH_TOKEN
  2. agentpassgrant issued · secret not in context
  3. modelsees: allowed / denied. never the token.
  4. logs, traces, chat · redacted

You can paste a secret into a prompt once. Then it lives in the thread, the logs, the provider, the laptop. agentpass keeps those bytes out: the agent requests a capability, the vault performs it or refuses it, and the secret never enters the model.

The agent never reads it
The agent does not read the password to use the password. It does not get a decrypted blob "just for this turn." The vault holds the material. The agent holds a grant.
It cannot leak from chat
If the value never enters context, it cannot be echoed, logged, or trained on from the chat. A human password manager is built for a person looking at a field.
Early
You are already handing agents a .env and hoping the system prompt holds. The secret should never have been in the prompt.

The rest of the chain