---
title: Agent Management
summary: Autonomy levels, the version history behind every change, and templates for managing a fleet of agents day to day.
date: 2026-09-17
---

# Agent Management

Autonomy levels, the version history behind every change, and templates for managing a fleet of agents day to day.

Once you're running more than one or two agents (see [First Setup](/en/docs/getting-started/first-setup) if you haven't created your first), day-to-day management comes down to three things: how much you trust an agent to act on its own, how you track changes to its behavior, and how you start new ones without rebuilding from scratch.

## Autonomy levels

Every agent runs at one of three autonomy levels:

- **Supervised**: the agent proposes an action and waits for you to approve it before anything runs.
- **Semi-auto**: you approve the plan up front, then the agent executes it without further check-ins.
- **Autonomous**: the agent runs end to end with no approval step at all.

Start new or unproven agents on **Supervised** and move them to **Autonomous** once you trust their judgment on a given task type.

## Configuration history

Every configuration change to an agent (its standing orders, provider, tool access) is saved as a new version automatically, so there's always a record of exactly what an agent's configuration looked like at any point in time. This matters most for standing orders (see [Core Concepts](/en/docs/guides/core-concepts)), since they're the single biggest factor in agent quality.

## Templates

Rather than building every agent from a blank standing order, start from one of six built-in templates (Customer Support Agent, Research Assistant, Content Creator, Lead Generation Agent, Personal Assistant, or Discovery Agent) and adjust from there.

## Related

- [Models](/en/docs/guides/models): assigning providers per agent
- [Security](/en/docs/guides/security): how tool access and credentials are scoped per agent
- [Advanced Configuration](/en/docs/guides/advanced-configuration): tuning standing orders and memory behavior further