Secrets stay out of the terminal
Agents need API keys. They don’t need to see them. How the vault hands out credentials one approval at a time.
Agents need credentials. They deploy previews, call staging APIs and run migrations against test databases. The usual answer is to paste a token into a .env file or an environment variable and hope nobody prints it.
Agents print things. That’s their job. And a terminal’s scrollback, a session transcript or a log file is a bad place for a production token.
Ask, don’t hold
ORRERION keeps credentials in a vault on your Mac. Agents don’t hold secrets; they ask for them by name. When an agent asks for STRIPE_TEST_KEY, its lane turns amber and ORRERION shows you exactly which agent wants which secret, for which task.
Say yes and the value is handed to that agent’s process. Say no and the agent is told it was refused. Either way, you decided.
Never in the terminal
Approved values go straight into the agent’s environment. They’re never typed into the terminal, never written to shell history and never shown in the conversation. If an agent tries to echo one, there’s nothing for it to read back from the scrollback.
A record of every request
Every request is logged: the agent, the task, the secret’s name, and whether you approved it. When something unexpected happens, you can see what had access to what, and when.
What we don’t do
- We don’t sync your vault to our servers. It stays on your Mac.
- We don’t auto-approve. Every release is a decision you make.
- We don’t pretend this replaces a real secrets manager for production. It’s about keeping agents on your laptop honest.
Read how it works in the docs, and how agents request secrets through the MCP server.
— The ORRERION team