Skip to content

Credentials Vault

The HSM-backed credentials vault enables agent-blind execution: agents submit tool calls, and Vargate looks up credentials from the secure vault and executes on their behalf. The agent never sees the credential value.

How It Works

Agent → Vargate Proxy → HSM Vault (fetch credential) → Execute Tool → Return Result
  1. You register a credential for a tool (e.g., Stripe API key)
  2. Agent submits a tool call for that tool
  3. If policy allows, Vargate fetches the credential from the HSM
  4. Vargate executes the tool call with the credential
  5. The result is returned to the agent
  6. The credential value is never logged or returned

Security benefit

Compromising the agent doesn't compromise your credentials. The agent never has access to secrets.

Endpoints

Register a Credential

POST /credentials/register
Content-Type: application/json

{
  "tool_id": "stripe",
  "name": "api_key",
  "value": "sk_live_abc123..."
}

The value is encrypted and stored in the SoftHSM2 vault. It cannot be retrieved — only used for execution.

Store securely

The credential value is write-only. You cannot read it back after registration. Keep a copy in your organization's secrets manager.

List Credentials

GET /credentials

Returns metadata about registered credentials (never the values):

{
  "credentials": [
    {"tool_id": "stripe", "name": "api_key", "registered_at": "2026-04-01T..."},
    {"tool_id": "gmail", "name": "api_key", "registered_at": "2026-04-02T..."}
  ]
}

Credential Status

GET /credentials/{tool_id}/status

Check if a credential is registered for a specific tool:

{
  "tool_id": "stripe",
  "registered": true,
  "name": "api_key"
}

Remove a Credential

DELETE /credentials/{tool_id}/{name}

Permanently removes the credential from the vault.

Access Log

GET /credentials/access-log

View a history of credential access events — which agents triggered credential usage and when.

Execution Modes

When a tool call is allowed by policy, the response indicates how it was executed:

Mode Description
agent_direct No credential registered. Agent handles execution.
vargate_brokered Credential fetched from HSM. Vargate executed the call.

The execution_mode field in the response and audit log tells you which mode was used.