AI Agents Guide

Safety Controls

Configure rate limits, action budgets, and resource scopes to ensure AI agents operate safely.

Estimated time: 15 minutes

Controls

Safety Control Types

Three layers of protection for AI agent operations.

Control request frequency and token usage.

rate_limits:
  requests_per_second: 10
  requests_per_minute: 100
  tokens_per_hour: 50000
  concurrent_sessions: 5

Limit operations per session or time period.

action_budget:
  per_session:
    api_calls: 50
    database_queries: 20
    file_operations: 10
  per_day:
    total_actions: 1000

Restrict access to specific resources.

resource_scope:
  allow:
    - "db/analytics-*"
    - "api/reporting/*"
    - "storage/reports/read"
  deny:
    - "db/*/write"
    - "api/admin/*"

Learn about automatic circuit breakers for anomaly detection.