Skip to main content

​
Purpose

In agentic systems, “who did this?” is often ambiguous. TealTiger treats identity as a first-class governance signal so you can:
  • apply least privilege policies
  • scope tool/model access
  • attribute cost
  • trace actions for audit and incident response
Identity is used in:
  • policy conditions
  • decision explanations (reason codes)
  • audit event evidence

​
Identity model (v1.1.0)

TealTiger supports identity at multiple layers:
  • Actor identity: the user or system initiating the request
  • Agent identity: the agent/service performing actions
  • Tool identity: which tool/connector executed the action (when applicable)
Identity is not only security. It is also cost attribution and reliability scoping.
For deterministic governance, provide stable identifiers:
  • actor.id (end user or calling service)
  • actor.type (user/service)
  • agent.id (agent instance or deployment)
  • agent.role (trusted/untrusted/system)
  • tenant_id / org_id (multi-tenant context)
  • environment (prod/staging/dev)

​
Why stable IDs matter

  • Policies can allowlist/denylist identities deterministically.
  • Audit events remain attributable across time.

​
Identity-based governance patterns

​
1) Least privilege tool access

  • Trusted service identities may access privileged tools.
  • Untrusted identities are restricted or denied.

​
2) Cost attribution

  • Attribute spend to actor/tenant.
  • Apply different budgets by identity class.

​
3) Reliability scoping

  • Apply stricter timeouts or degrade behavior for unknown identities.

​
Safe defaults

If required identity context is missing, policies should default to safe behavior. Common pattern:
  • missing identity → DENY with INPUT_MISSING_IDENTITY
Do not rely on user-provided free text identity. Treat identity as structured, verified metadata from your application.

​
Audit evidence

Identity should be recorded in the AuditEvent as structured metadata, not raw content. Recommended:
  • include stable IDs
  • avoid storing sensitive personal attributes
  • apply redaction rules to identity fields if needed
  • /policy/conditions-and-actions
  • /policy/reason-codes
  • /audit/audit-event-schema
  • /concepts/audit-and-redaction