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

# Deploy and set the mode

> Put an agent on a store, decide how far it may go there, and pause, run or stop it

A deployment is one agent on one store. An agent can run on many stores, and
each deployment has its own mode, its own status and its own history. An agent
starts its schedules, and can be chatted with, only once it is deployed.

Deploying and managing deployments needs `agent:write`.

## Deploy an agent

**1. Open the deploy dialog**

On the agent's page, choose **Deploy**. Or go to **Agents → Deployments →
Deploy agent** and pick the **Agent**.

**2. Pick the stores**

Choose one or more **Stores**. Stores it already runs on are not offered.

**3. Deploy**

**Deploy to 1 store** (or more). The deployment starts in the agent's own
mode, which is **Act** for every agent created in the dashboard.

To start on a store in a lower mode, deploy it, then change the mode on the
deployment's **Settings** before its first scheduled run.

## Find a deployment

**Agents → Deployments** lists every deployment: the agent, the store, its mode
and status, and when it last ran. Filter by **Agent**, **Store** or **Deployment
status**. An agent's own **Deployments** tab lists just its own.

A deployment's page has four tabs:

| Tab | What it shows |
| - | - |
| **Overview** | Agent, store, mode, status, when it started and last ran, open findings, its recent runs, and links to chat and to the store's changes |
| **Schedules** | The agent's checks as they run on this store, in store time: when each fires next and how the last one went. Edit them on the agent |
| **Runs** | This deployment's runs |
| **Settings** | The mode |

## Modes

The mode says how far an agent may go on its own, on one store.

| Mode | What it does |
| - | - |
| **Observe** | Watches and reports. Suggests nothing |
| **Recommend** | Proposes a specific change. You apply it |
| **Act** | Writes to Shopify itself. The Fraud agent tags orders and adds a risk assessment, which needs the store's `write_orders` permission. Every change is listed on the store's Changes tab, with a way to undo it |

Change it on the deployment's **Settings**, then save with the save bar.

> [!WARNING]
>
> Choosing **Act** needs the `agent:mode` permission, which only the Owner and
> Admin roles have by default. Anyone with `agent:write` can lower a deployment to
> Recommend or Observe.

The Fraud agent's order checks read what the buyer did on your storefront before
ordering. If storefront events are not reaching Glood, the deployment's Settings
says so; see [Set up storefront events](https://docs.glood.ai/platform/dashboard/getting-started.md#set-up-storefront-events).

## Deployment actions

The header of a deployment holds what its status allows:

| Action | When | What it does |
| - | - | - |
| **Run now** | Running | Starts a run immediately, the same way a schedule would |
| **Pause** | Running | Its scheduled runs stop on this store until resumed. A run already in progress finishes |
| **Resume** | Paused | Its schedules fire again from the next tick |
| **Stop** | Running or paused | Takes it out of service on this store, for good, until someone restarts it. Its findings, memory and history are kept |
| **Restart** | Stopped | Puts it back in service on this store |

Pause is the way to stop one store's checks while the agent keeps running
elsewhere. To stop the agent everywhere, [archive it](https://docs.glood.ai/platform/agents/index.md#archive-an-agent).


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