Guardrails 2.0 brings policy-as-code, streaming enforcement, and per-tool scopes to every Nexora agent.
A safer default
Guardrails 2.0 ships with sensible defaults and lets platform teams express policy as code that product teams inherit automatically. Enforcement now runs inline on every model output, tool call, and retrieved document — before a single token reaches your users.
What's new
- Policy-as-code: version allow and deny rules in YAML alongside the agent they protect.
- Streaming enforcement: validate and redact tokens as they stream, with no added round-trip.
- Per-tool scopes: grant each tool the narrowest set of arguments and destinations it needs.
- Inline PII redaction: strip sensitive spans before they are ever logged or shown.
Migrating
Existing policies keep working, so you can upgrade at your own pace and adopt the new primitives one agent at a time. Every blocked action is captured as a replayable trace, letting you tune thresholds against real traffic instead of guesswork. Guardrails 2.0 is available today on every paid plan.