# Agents

**Agents** is where you shape how Productmind's AI agents work in your workspace: the templates
and instructions they follow, the skills they can load, and, where automations are available,
agents of your own. Every change applies to this
workspace only, except the model provider, which is your organization's.

<Availability permission="Agents access on the Roles tab, under Administrative access. The Owner, Admin, Power User and Developer roles have it." />

:::note
Most of **Agents** configures Productmind's AI agents, their templates and skills. The
agent that builds coding sessions on your own machine is a different thing: the Productmind agent,
the `pm agent` command of the [Productmind CLI](/user-guide/tools/productmind-cli). To download and
install it, choose the **Productmind CLI** tab, the last one, or **Connect your Productmind agent**
in your profile menu.
:::

## Key concepts

**Template.** A document an agent follows: either the shape of something it writes, such as a
feature specification, or the instructions for a job it does, such as a
[coding session's](/user-guide/sessions/overview) plan.
See [Templates](/user-guide/engineering/agents/templates).

**Skill.** A named procedure an agent loads when it needs it, such as how to create a feature
your way. You can run a skill yourself in the PM agent by typing **/** and its name. See
[Skills](/user-guide/engineering/agents/skills).

**Workflow skill.** The instructions an external AI assistant, such as Claude Code, gets when it
works on your tasks through Productmind's MCP server. See
[MCP Workflow Skill](/user-guide/engineering/agents/mcp-workflow-skill).

**Definition.** An agent of your own, with its own instructions, model and tools, that an
automation can run. See [Definitions](/user-guide/engineering/agents/definitions).

**Model class.** A tier of model capability (**Quick**, **Standard**, **Top** or **Ultra**) and the
ordered list of models it runs on. Productmind keeps a list for every tier, and you can replace it
with your own for this workspace, and lower the effort coding sessions use with each model in it.
See [Model classes](/user-guide/engineering/agents/model-classes).

**Model provider.** Your organization's own Anthropic, OpenAI or LiteLLM key, which agents and
coding sessions run on instead of Productmind AI credits. See
[Model provider](/user-guide/engineering/agents/model-provider).

**Default, standard and custom.** Productmind ships a default for every template and a set of
standard skills and agents. You can edit any of them for your workspace, and restore the default
at any time. Until you edit one, your workspace gets Productmind's improvements to it
automatically.

## The tabs

In the sidebar, under **Engineering**, choose **Agents**. The tabs are listed down the left side
of the page, with **Overview** at the top, then the other tabs in four groups: **Models**,
**Instructions**, **Automations & MCP**, and **Local agent**. Choose a tab to open it beside the
list.

**Overview** opens first. It shows a card for each tab, in the same groups, with the tab's name
and one sentence about what you do there. Choose a card to open its tab. A tab your plan doesn't
include shows a lock, on its card and in the list.

This table lists the tabs below **Overview** in the order the page shows them.

| Tab | What you do there |
| --- | --- |
| [Model classes](/user-guide/engineering/agents/model-classes) | Choose which models each tier runs on in this workspace, the order they fall back in, and each model's effort level |
| [Model provider](/user-guide/engineering/agents/model-provider) | Connect your organization's own model key, and choose whether agents, coding sessions and developers' local agents use it. Appears when bring your own model key is available to your organization. |
| [Templates](/user-guide/engineering/agents/templates) | Edit the templates and instructions agents follow, and choose the model and reasoning effort for each job |
| [Skills](/user-guide/engineering/agents/skills) | Edit the standard skills and write your own |
| [Definitions](/user-guide/engineering/agents/definitions) | Create agents of your own, or customize the standard ones, for your automations to run. Appears when automations are available to your organization. |
| [MCP Workflow Skill](/user-guide/engineering/agents/mcp-workflow-skill) | Edit what external AI assistants are told about working on your tasks. Appears when it is available to your organization. |
| [Productmind CLI](/user-guide/tools/productmind-cli#installing-the-cli) | Nothing to set: the commands that download, install and connect your own Productmind agent, the `pm agent` command of the Productmind CLI, and whether your agent is connected. The **Local agent** group holds only this tab. |

## Permissions

| To | You need |
| --- | --- |
| Open **Agents** and change anything on it | **Agents** access (under **Administrative access** on the **Roles** tab) |
| Change or reset a tier's models and effort levels on **Model classes** | An **Owner** or **Admin** of your organization, the workspace's owner, or one of its admins |
| Connect, change or remove the organization's model key on **Model provider** | The **Owner** or **Admin** role |

The **Owner**, **Admin**, **Power User** and **Developer** roles have **Agents** access.
**Product Manager**, **Viewer** and **Customer** don't see **Agents** in the menu. Everyone's
agents still use what you set here. See
[Roles and permissions](/user-guide/reference/roles-and-permissions).

## Related topics

- [Automations](/user-guide/engineering/automations)
- [Applications](/user-guide/engineering/applications)
- [AI assistants (MCP)](/user-guide/tools/ai-assistants)
- [Productmind CLI](/user-guide/tools/productmind-cli)
