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

Getting Started

Connect one route, observe one incident timeline, and enforce one guardrail boundary first.

docs.cyiro.comproduction-ready guidance

Step 1: Connect runtime events

Send request/response lifecycle events with route, model, latency, and cost metadata so Cyiro can build operational traces.

  • Example: POST /v1/ingest/runtime with route_id, model_name, latency_ms, token_count
  • Example: Include policy_profile_id to map events to guardrail packs
  • Example: Use SDK helpers for automatic batching and retry logic

Step 2: Activate guardrail policies

Enable baseline runtime policies for unsafe output, latency spikes, and cost anomalies before expanding to custom rules.

  • Example: Enable unsafe_output_detection with confidence threshold 0.85
  • Example: Set latency_spike_threshold to 2000ms for critical routes
  • Example: Configure budget_variance_alert at 15% over baseline

Step 3: Attach provider watchers

Monitor pricing, status, and release-note surfaces for the providers your active routes depend on.

  • Example: Add Anthropic watcher for claude-3-opus route dependencies
  • Example: Configure AWS Bedrock watcher for titan-express-v1 endpoints
  • Example: Set watcher poll frequency to 5 minutes for production routes