Can an AI agent read my secrets?

No. Agents can list, create, update, and delete secrets by name, but there is no verb that returns a value to an agent. Plaintext is decrypted server-side only, for humans, on an explicit reveal.

How are secrets encrypted?

Each value is sealed with AES-256-GCM using a key derived for the secrets namespace. The plaintext is never stored: only the encrypted envelope and a masked last-4 preview.

Who can see a workspace's secrets?

Only members of that workspace, gated by the secrets permission. Non-members can't even tell a secret exists, and every reveal is written to the audit log.

Workspace

A vault for your keys.

Store API keys, tokens, and passwords: encrypted, never shown to agents.

Use for Free See pricing

What you get

Encrypted at rest

Every secret is sealed with AES-256-GCM under a secrets-namespace key. A database dump alone is inert.

Never shown to agents

Agents manage secrets by name but can never read a value: plaintext stays server-side.

Reveal on demand

Values are masked by default. Reveal is an explicit, audited, rate-limited action.

Workspace-scoped

Secrets live in a workspace and are gated by membership: share a workspace, share its secrets; lose membership, lose access.

One clean surface

Manage them from the app, the CLI (`idapt secret …`), the SDK, or your agent: one v1 API.

How it works

1
Add a secret
Name it (OPENAI_API_KEY), paste the value: it's encrypted before it ever reaches the database.
2
Use it by name
Reference the secret from the CLI, SDK, or an agent tool. The value is resolved server-side, never exposed.
3
Reveal or rotate
Reveal the plaintext when you genuinely need it (audited), or rotate it in place to retire the old value.

Frequently asked

Ready to try it?

Start free with a welcome credit. No credit card.

Try now Read the docs

Read the docs

Pair with