All videos

AgentsRoom Secret Manager: your AI agents use an API key without ever seeing it

Filmed in the real app: an agent fails a payment test because the Stripe test key is missing. The key is saved once in the Secrets tab, typed into a masked field, then handed to every agent of the project by its name through one project environment variable. The restarted agent and a new agent both pass the test, the key never shows up in their chat, and an agent asked which secrets it can use answers with names only.

Read the full feature page

Transcript

Your AI agents can use your API key without ever seeing it. This is AgentsRoom, the desktop app where your AI coding agents work side by side, on Claude Code, Codex and more. This agent has to run the payment test, and it needs the Stripe test key.

Without it, the test fails. The quick fix is pasting it into the chat. Then the key sits in the conversation, and maybe in your repo.

Instead, I open Commands and connections, then Secrets. New secret: a name, and the key, typed into a masked field. Save.

The key is encrypted by your system keychain, and the list only shows its name and a reference. Then, in Project settings, AI providers, one environment variable hands that reference to every agent of the project. This agent was already running, so I restart it.

Same request as before. This time the test passes, and the key never shows up in the chat. A new agent from the quick launch menu gets it too, and passes on the first try.

Ask an agent which secrets it can use: it answers with names. No tool ever hands it a value. The Secret Manager is in every project, under Commands and connections, Secrets.

The link is in the description.