Some websites have no API, and some only work once you're signed in: a supplier portal, a government site, an old system the company still lives in. The Data Visions Browser Extension lets your module's agent work in the person's own Chrome or Edge, on sites they're already signed in to, while they watch. No passwords handed over, no robots running in the dark.
How it gets switched on
Three things have to be true before an agent can use the browser:
| Who | What | Where |
|---|---|---|
| A Workspace Admin | Turns the browser on for the workspace | Profile menu › Settings › Browser extension (the switch at the top) |
| The person | Installs the extension and clicks Connect this browser | Same page: Download, install, Connect |
| Their role | Includes Let assistants work in their browser | System settings › Users and roles (administrators have it already) |
When the workspace switch is on, agents can use the browser. A Workspace Admin can still keep one agent off it: untick Browser for that agent on System settings › Agents.
As the creator, you build your agent to be good at browsing: tell it in its instructions when to use the browser, and give it room for more steps (see the developer snippet).
What the agent can do
| It can | For example |
|---|---|
| Open a page | Go to the supplier's order page |
| Read the page | See the headings, buttons and fields, or pull the text out in chunks |
| Find things | Locate "Invoice total" or every link in a menu |
| Click | By the button's words, or by its exact spot on the page |
| Type | Fill a box, or read and fill a whole form in one go |
| Screenshot | Show the person what it sees (the agent itself can't look at pictures) |
| Wait | Until something appears on the page |
Long pages go into a scratchpad the agent can search, so it doesn't drown in text. The scratchpad clears when the agent's turn ends.
Safe by design
- It's the person's browser. An agent working for one person can never reach anyone else's browser, even in the same workspace.
- Its own tabs only. The agent works in a tab group called Data Visions Agent, with a coloured border, and never touches the person's other tabs.
- Sign-ins stay with the person. If a site wants a login, the agent stops and asks the person to sign in, then carries on when they say so.
When to use it, and when an API is better
| Use the browser when | Use something else when |
|---|---|
| The site has no API | The site has an API: call it from your module (see Outside hosts and Python packages) |
| The person is already signed in there | It's your own module's data: use an agent tool |
| Someone is at the keyboard to watch | It runs overnight or on a schedule: the browser is only there while the person's browser is open |
| Seeing it happen helps | It's another Data Visions module: use Services |
Rule of thumb: the browser is slower and uses more AI than a direct call. Save it for the jobs that truly need a page.
Ask the Module Builder
// For developers: module.json, ai.agent
"max_iterations": 50,
"config": { "browser_extension_enabled": true }Good to know
- Chrome and Edge only. Firefox and Safari aren't supported.
- One browser at a time. With the extension on two computers, the agent uses the one that checked in most recently.
- Running out of steps? The agent stops, says how far it got, and asks whether to carry on. That's on purpose: start at 50 steps rather than a huge number.
- Some sites block page scripts. The agent's clicking and typing still work there.
- More on teaching an agent to browse: Agents in the browser: the browser extension.