Docs center

Self-hosted docs for Guardrails for AI.

Technical guidance for observability, guardrails, permissioning, and automation in one first-party documentation surface.

Self-hosted docs

Agentic Ops

Operate multi-step agent workflows with policy enforcement and runtime reliability controls built in.

docs.cyiro.comproduction-ready guidance

Execution graph control

Trace each tool request, policy check, and model handoff as one coherent run with shared context.

  • Example: Agent run shows tool requests with input/output logging
  • Example: Policy checks displayed with pass/fail status and reasoning
  • Example: Model handoffs tracked with context preservation details

Runtime governance

Combine permission boundaries, policy packs, and fallback automation to keep agent workflows safe at scale.

  • Example: Permission boundaries enforced at each tool invocation
  • Example: Policy packs applied consistently across all agent runs
  • Example: Fallback automation triggered when primary workflows fail

Run health scoring

Continuous evaluation of agent execution health and reliability.

  • Latency metrics: Response time analysis for each step
  • Error rates: Failure tracking across tool invocations
  • Policy compliance: Guardrail adherence monitoring
  • Context preservation: Data consistency validation
  • Resource usage: Token and compute consumption tracking

Tool-call governance

Fine-grained control over agent tool usage and permissions.

  • Pre-approval requirements: Mandatory review for sensitive tools
  • Context validation: Input/output safety checks
  • Rate limiting: Usage quotas and throttling
  • Audit trails: Complete logs of all tool invocations
  • Fallback mechanisms: Alternative tools for failed operations

Human override points

Strategic intervention opportunities in agent workflows.

  • Approval gates: Manual sign-off for critical decisions
  • Exception handling: Human review of policy violations
  • Context enrichment: Additional information for complex scenarios
  • Workflow modification: Dynamic adjustment of execution paths
  • Escalation procedures: Hierarchical review processes

Execution replay and review

Post-execution analysis and debugging capabilities.

  • Step-by-step replay: Detailed reconstruction of agent runs
  • Policy evaluation: Retrospective guardrail analysis
  • Performance metrics: Latency and resource usage breakdown
  • Context inspection: Data flow and state examination
  • Comparative analysis: Side-by-side run comparisons

SLA policy matrix

Service level agreements for agent operations.

  • Response time targets: Maximum latency for agent executions
  • Success rate requirements: Minimum completion percentages
  • Policy compliance levels: Guardrail adherence expectations
  • Fallback activation thresholds: Conditions for failover triggering
  • Notification requirements: Alerting for SLA violations