# Browser Agent

> Hand Seovyn browser jobs that have no API, like submitting a sitemap or checking a post renders. Pilot, Seovyn's Chrome extension, does them in your own browser and asks before anything sensitive.

Some SEO work has no API: submitting a sitemap in Search Console, checking that a post renders, tidying a draft in your CMS. The **Browser Agent** lets you hand those jobs to Seovyn. **Pilot**, Seovyn's Chrome extension, carries them out in your own Chrome, in the tab you pick, signed in as you, and asks before anything sensitive.

> [!NOTE] In beta
> Pilot isn't listed in the Chrome Web Store yet. To try it, ask in the chat or write to support@seovyn.com.

## Connect Pilot

:::steps
### Install Pilot in Chrome

Add the Pilot extension to Chrome and open its side panel.

### Sign in with Seovyn

In the side panel, choose **Sign in with Seovyn**. A Seovyn tab opens and shows what Pilot will be allowed to do for your workspace:

- carry out the browser tasks you give it,
- ask before it submits, buys or deletes anything,
- use this workspace's Browser Agent budget.

Choose **Connect Pilot**, then head back to the side panel. If you're signed in to the wrong organization, choose **Switch** first.

### Check it's online

On the **Browser Agent** page in the dashboard, the runtime shows **online** once Pilot is connected.
:::

You can sign out of Pilot at any time from its menu. To pair by hand instead, the Browser Agent page has **Pair extension**, which gives you a workspace token and the agent server address to paste into Pilot's settings.

## Give it a task

On the **Browser Agent** page, describe the task in plain words under **What should Pilot do?** Pilot plans the clicks. For example:

- Publish my latest approved article in WordPress and check it renders.
- Submit the sitemap in Google Search Console.
- Check the partners page for broken backlinks.

Choose the **permissions** (below), optionally a model, and **Create task**. Then, in Pilot's side panel, pick the task and connect it to the tab you want it to work in.

## Permissions

| Mode | How it behaves |
|---|---|
| **Supervised** (the default) | Reading pages and scrolling need no approval. Before Pilot changes anything on a site, or does anything sensitive like submitting a form, buying or deleting, it shows you exactly what it's about to do and waits. An approval request expires after five minutes |
| **Auto · time-limited** | Runs unattended for the window you choose (15 minutes, 1, 4 or 8 hours), even for sensitive steps. When the window ends, it asks again before anything sensitive |

> [!WARNING]
> Use Auto only on sites and tasks you trust. Pilot works in your own signed-in browser, so it can do whatever you can do on that site.

## Follow a task

**Recent tasks** lists the latest 40 tasks with their status, approvals and how many tokens each used. Filter by **Running**, **Needs attention** (waiting for an approval) or **Finished**. The totals at the top cover all your saved tasks: tasks, running now, tokens used and estimated spend.

## Budget

Every plan includes a monthly Browser Agent budget for the AI models Pilot uses:

| Plan | Budget a month |
|---|---|
| Free | $1 |
| Pro | $5 |
| Scale | $20 |

Before each model call, Pilot sets aside an estimate from the budget, and settles the real cost when the call returns. A task pauses before it would go over the budget, so there's never an overage charge; it resumes when the budget resets. **Plan & usage** shows what's used and held, and what each task cost. See [Plans and usage](/docs/plans-and-usage).