Skip to main content
An identity represents a single AI agent (or agent instance) in your project. It’s how the gateway knows who is making a request and what metadata to attach to policy evaluations.

​
What’s in an identity

​
Credentials

Each identity has a credential — a secret string the agent uses to authenticate with the gateway. MCP clients include it as a bearer token in the Authorization header:
Credentials are:
  • Securely stored — credentials are hashed and never stored in plaintext
  • Single-use display — the secret is shown once when created, then never again
  • Revocable — you can revoke a credential at any time from the console
  • Expirable — optionally set an expiration date
Store credentials securely. If lost, generate a new credential — the old secret cannot be retrieved.

​
Labels

Labels are key-value pairs attached to an identity. They’re available in policy evaluation as identity.labels.
Common label patterns:
  • team: support / team: engineering / team: data
  • env: production / env: staging
  • role: agent / role: service
  • vendor: openai / vendor: anthropic
Labels are flexible — use whatever makes sense for your policy structure.

​
Allowed tools

By default, an identity can see all tools registered in the project. You can restrict this by setting an allowed tools list.
Task-level allowed tools override identity-level restrictions. If a task defines its own tool list, it takes priority.

​
Identity in policy context

During policy evaluation, the identity is available as:

​
Next steps

Create identities

Step-by-step guide to creating identities in the console.

Tasks

Learn how tasks add scoped context on top of identities.