> ## Documentation Index
> Fetch the complete documentation index at: https://docs.sphinx.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Inspect and govern learned knowledge

> Review the knowledge agents use, trace claims to sources, and correct errors through a reviewable workflow.

Sphinx makes learned context available as readable knowledge. Your team can inspect what an agent consulted, check the supporting material, and make a targeted correction when the context is wrong.

## Inspect the knowledge behind a claim

Start in [Browse Knowledge](https://docs.sphinx.ai/knowledge-base/browse.md). Read the relevant page, follow its related concepts, and examine the citations for the claim you are checking. With MCP, ask your agent to read the complete pages it finds and include source links in its answer.

This makes the retrieved context inspectable. An answer still needs review against the task and the original evidence.

## Control how knowledge changes

| Control | How to use it |
| - | - |
| **Proposed changes** | Inspect the page differences and accept or reject an update before applying the proposal. |
| **Controversies** | Keep competing claims visible until the responsible owner supplies a resolution. |
| **Page lock** | Prevent Sphinx updates to a page that needs to remain fixed. |
| **Audit** | Inspect workspace events and activity when investigating changes or usage. |

See [Review and accept knowledge changes](https://docs.sphinx.ai/knowledge-base/review-changes.md) for the decision workflow.

## Investigate an incorrect answer

1. Record the question, answer, and sources or context the agent used.
2. Determine whether the issue comes from missing knowledge, an incorrect rule, an unresolved disagreement, or the agent's interpretation.
3. Submit a precise correction through [ingest](https://docs.sphinx.ai/knowledge-base/ingest.md) or [controversy resolution](https://docs.sphinx.ai/knowledge-base/controversies.md). Include scope, exceptions, and supporting evidence.
4. Review and accept the proposed correction.
5. Rerun the original question and related questions to check the result. Add a [benchmark](https://docs.sphinx.ai/benchmarks.md) to catch the same failure in future evaluations.

For example, a rule that incorrectly applies an exception to every case can cause an agent to reject valid evidence. Correct the rule's scope, then test cases both inside and outside the exception.

## Plan your deployment

Choose [MCP](https://docs.sphinx.ai/mcp/overview.md) or the [Private Gateway](https://docs.sphinx.ai/private-gateway/overview.md) according to how you want agents to access context and inference. Work with Sphinx to confirm the data handling, model access, and audit requirements for your organization's deployment.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.