> For the complete documentation index, see [llms.txt](https://surecloud.gitbook.io/surecloud-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://surecloud.gitbook.io/surecloud-docs/documentation/gracie/personas-skills-and-agents/agents.md).

# Agents

Define an AI agent once and attach it to workflow stages, transitions, and buttons so Gracie can run automatically.

## Agents

The **AI Agent Library** lets administrators define reusable AI agents and attach them to workflows.

This moves Gracie beyond chat.

Instead of waiting for a user to ask, an agent can run automatically when the workflow reaches the right point.

{% hint style="info" %}
Think of an agent as a deployed worker. Skills are invoked in the moment. Agents are deployed into a process.
{% endhint %}

### What an agent is

Each agent is a reusable configuration made up of a few core settings:

{% stepper %}
{% step %}

#### Name and instructions

Give the agent a clear name and a plain-language instruction. This tells the agent what it should do.
{% endstep %}

{% step %}

#### Skills

Choose from the available [Skills](/surecloud-docs/documentation/gracie/personas-skills-and-agents/skills.md). This controls which capabilities the agent can use.
{% endstep %}

{% step %}

#### Internet access toggle

Turn internet access on or off. Some agents should only use internal data, while others may need to reach external sources.
{% endstep %}

{% step %}

#### Permission assignment

Assign the role and hierarchy position the agent will use. An agent can only do what that assigned access allows.
{% endstep %}
{% endstepper %}

### When to use an agent

Use an agent when the work is repeatable and the trigger is stable.

Good examples include document review, onboarding checks, privacy screening, and workflow hand-offs.

If the task depends on a one-off question in the moment, use chat instead.

### Managing agents

The Agent Library in **Settings** lets admins manage agents for the tenant.

From here, you can:

* Create a new agent from scratch.
* Edit an existing agent's instructions, skills, or permissions.
* See where each agent is deployed across your workflows.
* Disable an agent temporarily without deleting it.

### Running agents on workflows

Once defined, agents can be attached to workflow behaviour in three places:

{% tabs %}
{% tab title="Stage logic" %}
Run an agent automatically when a record enters a stage. For example, when a SOC 2 report reaches a review stage, the agent can summarise it and populate key fields.
{% endtab %}

{% tab title="Transition logic" %}
Run an agent as part of a transition between stages. This is useful for checks or updates that should happen before the transition completes.
{% endtab %}

{% tab title="Button press" %}
Expose an agent behind a button on a record so users can run it on demand without opening chat.
{% endtab %}
{% endtabs %}

### Common use cases

Typical use cases include:

* **Third-party monitoring** — agents that check supplier trust centres and flag changes.
* **Document processing** — agents that read uploaded SOC 2 or ISO documents and produce structured summaries.
* **Questionnaire automation** — agents that create, send, and follow up on questionnaires.
* **Workflow decisions** — agents that review a record, apply a rubric, and prepare the next action for approval.

{% hint style="success" %}
Agents are built from the same skills catalogue that powers Gracie in chat. This helps keep behaviour consistent across manual and automated use.
{% endhint %}

See [Personas, Skills and Agents](/surecloud-docs/documentation/gracie/personas-skills-and-agents.md) for the full model.

### Reviewing agent activity

Each agent run is visible through the record and through Gracie. You can review:

* Whether the agent succeeded or failed.
* What output it produced.
* The reasoning it followed.

This follows the same transparency model used in Gracie chat.

### Permissions and control

Agents do not bypass permissions. They run within the access granted to the assigned role and hierarchy position.

Admins stay in control of which agents exist, where they are deployed, and whether they are enabled.

For broader controls around Gracie usage, approvals, and auditing, see [Governance & Controls](/surecloud-docs/documentation/gracie/governance-and-controls.md).