Guardrails
==========

Guardrails are safety and compliance policies applied to agent inputs and outputs. They intercept content before it reaches the agent or before the agent's response is returned to the user, and either block, mask, or flag content that violates the policy.

.. image:: /images/foundry-guardrails-list.png
   :alt: Guardrails list page
   :align: center
   :width: 100%

|

Because guardrails are Foundry resources, they are created once at the workspace level and can be applied to any number of agents. A single guardrail built by your compliance team can protect every customer-facing agent in the workspace.

OneByZero Neo supports two guardrail types, accessible from tabs on the Guardrails page:

* **In-House** — prompt-based guardrails powered by Neo's own evaluation engine. You define a validation prompt and configure what constitutes a pass or fail result.
* **Bedrock** — integration with AWS Bedrock Guardrails, which provides managed content filtering, denied topics, word blocklists, PII protection, and contextual grounding checks.

Switch between types using the **In-House** and **Bedrock** tabs at the top of the Guardrails page.

**Managing guardrails**

Both guardrail types support:

* **Search** — find guardrails by name using the search bar.
* **Tagging** — apply tags to organise guardrails by department, use case, compliance framework, or environment (e.g. ``production``, ``banking``, ``pii-protection``).
* **Versioning** — each guardrail displays its current version. New versions can be created while previous versions remain accessible.
* **Export / Import** — guardrail configurations can be exported and imported, enabling sharing across workspaces or promotion across environments.
* **Enable / Disable** toggle — controls whether the guardrail is active and can be applied to agents.

**In this section:**

.. toctree::
   :maxdepth: 1

   in_house
   bedrock