SecretBar — Agent-safe secrets for Mac - Your agent knows the name. Never the value.
by•
SecretBar is a free, open-source Mac app and command line tool that keeps secrets safe from AI agents. Passwords, tokens and API keys live in the Secure Enclave. The agent asks for a secret by name, you type it in a hidden macOS dialog, and the value goes only into the process that needs it. Output comes back masked as ***. Touch ID to read, post-quantum encryption, a Claude Code hook, time-boxed leases for unattended agents, Spotlight and Siri search. No cloud, no dependencies.

Replies
Hi Product Hunt! 👋
I'm Andrew, maker of SecretBar (and of Gantry, which some of you met in June).
I work with AI agents in the terminal all day: Claude Code, Codex. Sooner or later every agent needs a database password or an API token. The usual paths are bad: paste the value into the chat, or let the agent
cat .env. Either way the secret ends up in a transcript, a log or a model provider's servers. So I built the tool I wanted: the agent works with names of secrets, and never with the values.How it works:
secret ask NAMEopens a macOS dialog, so you type the value there, not in the chat.secret run --mask NAME -- cmdputs the value only into the environment ofcmdand replaces it with***in the output, also in base64 and URL encoding.secret get,cat .envandenvbefore they run.It's free and open source (MIT), one
brew install secretbar/tap/secretbar, no cloud and no dependencies beyond macOS.I'd love your feedback, especially from people who run agents on real projects. How do you handle secrets with agents today? What would make you trust a tool like this?