When an agent goes wrong the question is always the same: what did it do, was it authorised, can you prove it.

Only 21% of companies have a mature governance model for autonomous agents.

Transient wraps your agent process and enforces a permission policy before execution. Define what the agent is allowed to do. Anything outside that gets blocked. Every action gets a signed, tamper-evident receipt the agent can't alter.

pipx install transient-trace transient-trace wrap install git curl pip uv sudo --auto-rc source ~/.zshrc transient-trace run claude

Start in audit mode to see what your agent is actually doing. Switch to strict when you want to enforce.

No code changes. Works with Claude, any CLI tool, anything you can wrap.

Alpha. Keen to hear from anyone deploying agents in production.

github.com/james-transient/transient