Guardrail Composition

Compose complex guardrail logic using chains to create sophisticated validation workflows.

Chaining Strategies

import "github.com/MitulShah1/openai-agents-go/guardrail"

chain := guardrail.NewChain().
    // Add guardrails
    Add(security.NewPII(security.WithTripwire(true))).
    Add(moderation.NewProfanity(moderation.ProfanityConfig{Tripwire: true})).

    // Choose strategy
    WithStrategy(guardrail.Sequential). // Stop at first failure (Short-circuit)
    // OR
    // WithStrategy(guardrail.Parallel). // Run all concurrently
    // OR
    // WithStrategy(guardrail.StopOnFirstPass). // First pass wins

    Build()

// Use the chain as a single guardrail
runner.Run(..., agents.WithGuardrails([]agents.Guardrail{chain}))

Async & Timeouts

Apply timeouts to guardrails to prevent slow validation from blocking execution:

// Timeout after 500ms
timedGuard := guardrail.WithTimeout(slowGuardrail, 500*time.Millisecond)

// Timeout with graceful degradation (warns instead of failing)
gracefulGuard := guardrail.WithTimeoutGraceful(slowGuardrail, 500*time.Millisecond)

Metrics

Track guardrail performance with built-in metrics collection:

metrics := guardrail.NewInMemoryMetrics()
monitoredGuard := guardrail.WithMetrics(myGuardrail, metrics)

// ... run agent ...

stats := metrics.GetStats("my_guardrail")
fmt.Printf("Avg Latency: %v", stats.AvgDuration())

Legacy Chaining (Simple List)

You can also pass a simple list of guardrails to the runner. This is equivalent to a Sequential chain.

guardrails := []agents.Guardrail{
    security.NewPII(security.WithTripwire(true)),
    security.NewSecrets(security.SecretsConfig{Tripwire: true}),
}