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

# Know your agent

> An inventory of what is deployed, with an owner, a provenance and a model binding. The question every regulator opens with.

Before anything can be governed it has to be known. The first question in every
examination, every incident review and every security questionnaire is some
version of: **what have you got running, and who owns it?**

Most organisations discover two things when they answer it honestly. The list is
longer than expected, because "agent" turns out to include scheduled automation
that now calls a model. And a meaningful fraction has **no owner at all**: the
person who built it left, the team reorganised, and it has been running quietly
since.

That second finding is usually worth the exercise on its own.

## Agents name themselves

```http theme={"system"}
POST /v1/agents
POST /v1/agents/resolve
```

Registration is idempotent by name within a workspace, so it is safe on every
boot, and an agent appears in the inventory the first time it speaks rather than
when somebody remembers to add it. An inventory that depends on a human step is
an inventory that is wrong within a month.

```python theme={"system"}
rs.register("claims-orchestrator",
            owner="sarah.bennett@acme.example",
            org_unit="claims",
            tier="L2")
```

`owner` is a **person**. Not a mailbox alias, not a team. Everything the agent
later does walks back to this, and a distribution list at the end of that walk
is the absence of an owner written down.

## What the inventory holds

* **The agent**: name, owner, org unit, autonomy tier
* **Provenance**: which model, which version, reported by the SDK rather than
  typed into a form
* **Tools**: the MCP manifest it was seen with, hashed. See
  [Silent change](/concepts/silent-change)
* **Grants**: what it may currently do, and under which enforcement rung
* **Certification**: whether it was ever cleared to hold authority at all

## Autonomy tier

A coarse label, `L0` upward, describing how much an agent decides for itself.
It is documentation rather than enforcement: the bounds that actually stop
things are on the [grant](/concepts/grants). The tier exists so a human
scanning an inventory can see which agents deserve a closer look, which is a
different job from refusing an action.

## The gap this closes

Model inventories built by hand describe the deployment somebody remembered.
This one describes the deployment that is running, because it is assembled from
agents reporting themselves and from decisions already in the ledger.