Foundry
=======

The Foundry is the shared resource layer of OneByZero Neo. It is where administrators and workspace owners create and manage the building blocks that agents draw on — knowledge bases, integrations, guardrails, datasets, and evaluations.


Foundry resources are **created once and reused across multiple agents**. A single knowledge base can power ten different agents. A guardrail built by your compliance team can be applied to every customer-facing agent in the workspace. This centralised model keeps configurations consistent, simplifies updates, and reduces duplication.

The Foundry surfaces as a set of items in the left sidebar:

* **Knowledge bases** — document collections backed by OpenSearch (In-House) or AWS Bedrock. Agents use them for retrieval-augmented generation.
* **Integrations** — tools that agents can call at runtime: prebuilt In-House tools, Custom tools defined from OpenAPI specs, and MCP servers. The sidebar item is labelled **Integrations**.
* **Guardrails** — safety and compliance policies applied to agent inputs and outputs. Backed either by Neo's own prompt-based engine (In-House) or by AWS Bedrock Guardrails.
* **Datasets** — curated collections of test cases used as inputs for evaluations.
* **Evaluations** — evaluation runs that score agent responses against datasets using DeepEval (In-House) or AWS Bedrock Evaluations.


.. note::

   Foundry resources are also drawn on by :doc:`Digital Co-Workers </coworkers/index>` — Integrations and Agent Skills become a Co-Worker's skills, and guardrails and evaluations apply to a Co-Worker just as they do to any agent.

**In this section:**

.. toctree::
   :maxdepth: 2

   knowledge_bases/index
   integrations/index
   guardrails/index
   datasets/index
   evaluations/index