Hey Product Hunt — maker here. I build as VesperRun. I was tired of the same stupid leak. You give an agent a key so it can call GitHub or OpenAI. Ten minutes later the key is in the chat, the log, or a file the agent just wrote. The model did the work. The transcript kept the password. Veil is the tool I wanted on my own disk: • You save a name, not a message in a thread • You unlock a session • veil run starts one process with only the env it needs • You lock. Memory is wiped The agent can finish the job. It does not get a copy to quote back to you. It does not call a model. It does not phone home. It is not a platform. The child can still read os.environ — we do not pretend otherwise. We pretend nothing. We keep the key out of the one place that keeps getting it: the chat.

Veil-v1.0.0

Your agent can use the key. Chat never sees it.

Your agent can use the key. Chat never sees it.