Secrets vault
Encrypted, write-only values your AI agents can use inside their code sandboxes.
Some agent jobs need credentials — an API key for an insurer's portal, a token for an internal service. The Secrets vault stores them so agents can use them without anyone, the agent included, ever seeing the value.
How secrets work#
Open Settings → Secrets and choose Add secret. A secret is an environment value that agents can use inside their code sandboxes.
- Write-only. You can set, rotate and delete a secret, but never read its value back — not in the console, not through the API.
- Names, not values. Agents see a secret's name and use it in their sandbox; they never see what's inside it.
- Redacted output. If a value would appear in an agent's output, it's redacted.
Choosing which agents can use a secret#
Each secret has an access setting:
| Access | Meaning |
|---|---|
| All agents | Every agent in the organisation can use it |
| Selected agents | Only the agents you pick |
Prefer selected agents. A claims portal key belongs to the claims assistant, not to every agent in the workspace.
Rotating#
When a credential changes, or someone who knew it leaves, rotate it: update the value in place and every agent that uses it picks up the new one. You can also change a secret's description or its agent access without touching the value.
Through the API#
Secrets live under /docs/api/secrets:
GET /secretslists names and metadata — values are never returned;POST /secretscreates one;PATCH /secrets/{id}rotates the value or changes its description and agent access;DELETE /secrets/{id}removes it.
Create a secret for two agents:
POST /api/v1/secrets{ "name": "INSURER_PORTAL_KEY", "value": "…", "description": "Read-only key for the insurer claims portal", "access": { "mode": "selected", "agents": ["AGENT_ID_1", "AGENT_ID_2"] }}Listing needs secrets:view; creating, rotating and deleting need secrets:manage. Send secret values only from a trusted server, never from a browser.