Safety
Agency runs a team of AI agents on your computer, on your own Claude and Codex accounts. This page says plainly which programs Agency starts and what it does, and never does, with your sign-ins and keys.
Programs Agency starts
- Only the command-line tools themselves. Agency starts Claude Code and Codex from the copy it installed for you (in the
.agencyfolder in your home directory) or from theclaudeorcodexcommand already installed on your computer. It never looks inside another application, such as a desktop chat app, for a program to run. - Short background checks. Apart from your agents' sessions, Agency runs those same tools briefly to check that you are signed in and to read usage for the usage meters. These checks never start a conversation.
Your Claude and Codex sign-ins
- You sign in with the providers' own tools. Agency opens the official sign-in (
claude auth loginorcodex login) and shows its output in a window. Your sign-in stays with that tool, on your computer. - Agency never reads, stores or uploads your subscription credentials. To see whether you are signed in, it asks the tool, or reads the account name Claude records, never the token itself. Each agent starts with a short list of system settings that leaves out any provider tokens and keys Agency inherited.
- No pooling and no reselling. Your agents run on your computer under your own sign-in. Agency does not share your subscription with anyone, combine accounts or sell access to them.
- Keys you add are encrypted. API keys you add in Settings, such as an OpenRouter key, are encrypted with your operating system's credential protection.
Changes to this page
This page describes the current release and will change when its behaviour does.