Guardrail

class opik.guardrails.guardrail.Guardrail(guards: List[Guard], guardrail_timeout: int | None = None)

Bases: object

Client for the Opik Guardrails API.

This class provides a way to validate text against a set of guardrails.

classmethod from_stored_policies(names: Sequence[str] | None = None, guardrail_timeout: int | None = None) → Guardrail

Build a Guardrail from guardrail policies stored in your Opik workspace.

A policy is a named group of guards, stored in Opik. Referencing policies by name keeps the checks an application runs configurable without a code change. Policies the workspace applies unconditionally are always included, whether or not they are named here.

The guards of all retrieved policies are checked together, as one flat set: which policy a guard came from is not preserved.

Parameters:

  • names – Names of the policies to build the guardrail from. Every name must exist, a name that does not is an error rather than a check silently not running.

  • guardrail_timeout – Timeout in seconds for the guardrails backend calls.

Returns:

Guardrail running the guards of the retrieved policies.

Return type:

Guardrail

Raises:

  • opik.rest_api.core.ApiError – If the policies cannot be retrieved, for example when one of names does not exist in the workspace.

  • opik.exceptions.GuardrailPolicyError – If a policy holds a guard type this SDK version cannot run.

Example:

```python from opik.guardrails import Guardrail from opik import exceptions

guardrail = Guardrail.from_stored_policies(names=[“no_contact_information”])

try:

guardrail.validate(“Call me at 555-0123”)

except exceptions.GuardrailValidationFailed as e:

print(“Guardrail failed:”, e)

```

validate(text: str) → ValidationResponse

Validate text against all configured guardrails.

Parameters:

text – Text to validate

Returns:

API response containing validation results

Return type:

ValidationResponse

Raises:

  • opik.exceptions.GuardrailValidationFailed – If validation fails

  • opik.exceptions.GuardrailValidationError – If a guardrail cannot be evaluated (guardrails backend unreachable, timeout, or LLM judge provider failure). Guardrails fail closed, so this blocks the protected code path.