Hi PH — building toani Vault: AI agents that can use your passwords without seeing them

Hey everyone 👋

I'm Alex, founder of toani. First time posting here — wanted to introduce myself before we launch in a few weeks.

The short version of what we're building:

For the past year I kept hitting the same wall using AI agents. The second you give an agent a real password or API key to do anything useful — log into a dashboard, call an internal API — that secret is now in its logs, traces, and LLM context. Everywhere the agent goes, the secret follows.

So we built toani Vault: an open-source credential vault where the agent literally never gets the secret. It sends an intent ("fill the password field"), the credential is decrypted inside an Intel SGX enclave, and the plaintext never leaves. Everything MIT licensed — CLI, SDK, enclave.

A few things I'd love from this community as we prepare to launch:

- What's the worst credential-leak horror story you've seen with AI agents?

- Which agent workflows would you most want this in? (Browser automation? Customer support? CI?)

- Any feedback on the approach itself?

Repo's already up if you want to poke around: github.com/toaniAI/toani-vault

Happy to be here 🙏