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

# Tell SRE Agent how your organization works

> Write down how your team works, see what is waiting for review, and learn who can approve, edit, retire or flag a statement.

export const Plan = ({tier}) => <Badge color="blue">{tier} plan</Badge>;

Every team has rules that live in people's heads: "rollbacks need a second approver", "checkout failovers take about four minutes", "Dana prefers a phone call to a page at night". **Knowledge** is where you write them down, so that the right people can review them and nothing is kept as active before someone has read it.

<Plan tier="Free" />

The Knowledge page, the Knowledge section on a service and the MCP tools are on every plan. Drafting statements from investigations will belong to the Pro plan and above; nothing drafts statements yet, so Settings has no switch for it.

<Note>
  Today you can write, review, edit and retire knowledge, and every change keeps its history. SRE
  Agent does not use active statements in its answers yet. Nothing you write changes what it does
  until that arrives.
</Note>

## What a statement is

A statement is one plain sentence of up to 500 characters. It has three properties.

* **Kind**: a **Practice** is something your team does, an **Observation** is something SRE Agent measured, and a **Person note** is something about one person.
* **About**: the whole organization, a service, a team, or one person.
* **Status**: **Proposed** while it waits for a decision, **Active** once approved, and **Retired** when it is no longer true or was rejected. Retired statements stay with their history.

Statements are visible to everyone in the organization who can open the page, so never put a password or a key in one. A statement that looks like a credential is refused and the box is cleared. The refusal names the kind of secret it matched, never the value.

## Add and review knowledge

<Steps>
  <Step title="Open Knowledge">
    Click **Knowledge** in the sidebar, beside **Services**. Every role except guest can open it.
  </Step>

  <Step title="Add a statement">
    Click **Add knowledge**, choose what it is about, write one sentence and click **Add**. A member
    can write about the whole organization, a service or a team. Every role, a viewer included, can
    choose **About me** for a private note about themselves.
  </Step>

  <Step title="Review what is waiting">
    **Proposed** opens on the proposals you may decide, and a number on the tab counts them. Click
    **Show all proposals** to see the rest. **Approve** makes a statement active. **Reject** asks
    why, and keeps it in **Retired**.
  </Step>
</Steps>

Each row shows the statement, its kind and what it is about, whether it was flagged, **Learned** by SRE Agent or **Auto-approved**, and when it was last confirmed. A practice nobody has confirmed for 180 days is marked **Aging**. Open **History** on a row to see every event in order: who acted (a name, or SRE Agent), when, and the reason given.

An active statement offers **Edit**, **Retire** and **Flag**:

* **Edit** replaces the wording. The old version is retired and the new one goes through the same checks as a new statement.
* **Retire** asks why the statement is no longer true and takes it out of use.
* **Flag** is for any member who thinks an active statement looks wrong. It leaves use at once and waits in **Proposed**, flagged, until a reviewer reads it.

If the same statement is written again for the same subject, no second copy is made. The existing one is marked as confirmed again.

## Who can decide

Writing is a member action. Deciding belongs to the people who own the subject.

| Statement is about | Who may approve, reject, edit or retire it |
| - | - |
| The whole organization | Organization admins |
| A team | Members of that team and organization admins |
| A service | Members of the team that owns the service and organization admins |
| A person | The person it is about and organization admins, but only the person can edit it |

What happens when you add a statement follows from this. If you may decide the subject, it is active at once. Otherwise it waits in **Proposed** for someone who may. A statement that reads like an instruction to the AI, such as "ignore previous instructions", is stored but never used until a person approves it, whoever wrote it. History shows it as "Looks like instructions to the AI".

### Notes about a person

A person note is private. Only the person it is about and organization admins can see it, in the list, in its history and over MCP. Only that person can write or reword it, and administrators can approve, reject and retire it but never rewrite it. A note is deleted when the person's membership of the organization ends, or when they become a guest.

## Automatic approval

When the product proposes statements of its own, each source type earns a track record. After ten proposals in a row from one source were approved as written, an organization admin sees an **Accept** button for that source on the **Knowledge** card in **Settings**, on the **AI Providers** tab. Accepting lets that source's practices become active without review. Editing or rejecting a proposal restarts the count.

An admin can take it back at any time with **Revoke** and a reason, and rejecting one automatically approved practice takes its source back to review. For thirty days an admin sees a banner on the Knowledge page saying which source was revoked and why.

Observations that SRE Agent measures itself never need review, and what people write never counts toward a track record. A practice built from text in an alert or a card always waits for review. Accepting is refused while platform support is acting as your organization, so ask your own admin.

## Limits

* 20 new proposals per organization per day (UTC). A person who goes over is told to try again tomorrow. A statement that is active at once does not count, and editing never fails because of this limit.
* 2,000 active statements per organization. Retire some to make room.

## On a service

Every service page has a **Knowledge** section with the service's active and waiting statements, an **Add knowledge** form for members and **Review on the Knowledge page** when something is waiting. Deciding happens on the Knowledge page. See [Find and manage your services](/guides/get-started/services).

## Over MCP

Seven tools do the same as the page. Reads need the viewer role and writes the member role.

| Tool | What it does |
| - | - |
| `list_knowledge` | Lists statements, newest first. Filter by `status`, `scope`, `kind`, `mine_to_review` and `limit`. |
| `get_knowledge_item` | One statement with its history. |
| `add_knowledge` | Adds a statement for the organization, a service, a team or yourself. |
| `approve_knowledge_item` | Approves a proposal. |
| `reject_knowledge_item` | Rejects a proposal, with a reason. |
| `retire_knowledge_item` | Retires an active statement, with a reason. |
| `flag_knowledge_item` | Flags an active statement, with a reason. |

* Writes need a person, so connect with a user token. An API key can read but is refused every write, and it lists only statements the whole organization may read, never a person note.
* A token created with a lower role than its owner has is judged at that lower role.
* History says whether SRE Agent or a person acted and never names the person. The statement you add is kept out of the connection's own audit entry.

See [MCP tools](/api-reference/mcp) for how to connect a client.

## Audit log

Adding, approving, rejecting, retiring, flagging and editing a statement, and accepting or revoking automatic approval are recorded in the audit log as `knowledge.*` entries. They name the kind, subject and source, never the statement itself.


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