An AI browser agent that reads first and asks before it acts
Ask the agent to check a page, compare competitors or gather sources, and it drives a real browser in an isolated sandbox. Runs are read-only by default, interactive steps need your approval, and every run leaves steps, URLs and screenshots behind.
7-day trial with a card on file and 300 usage credits. Then $49.99/month unless you cancel.
Use cases
What agencies use it for
Competitor and pricing checks
Listing and website QA
Source-backed research
Report inputs
Safety model
Bounded by design
The browser agent is useful because it is constrained. These limits apply to every run.
| Control | Behavior |
|---|---|
| Default policy | Read-only. The agent can navigate, read and capture, but form submissions and other network writes are blocked. |
| Interactive runs | Require a specific approval before they start. |
| Sensitive inputs | Password and payment fields are refused in agent-driven typing. |
| Network guard | Private IP ranges and URLs with embedded credentials are blocked; redirects and requests are checked. |
| Bounds | Up to 12 steps and 7 minutes per run, inside a disposable sandbox. |
| Record | Each run keeps its steps, visited URLs, findings and screenshots for review. |
Beyond one-off runs
Schedules, logins and code
Scheduled checks
LiveVault-backed logins
Needs your accountApproved code runs
LiveLimits
Where the browser agent stops
- No single sign-on, multi-factor codes or payment flows.
- Logged-in sessions are short-lived and tied to the stored credential version, not kept indefinitely.
- File uploads and rich download handling are limited.
- It is not a substitute for direct ad, email or social platform integrations.
- Revoking a stored login clears its sessions immediately.
- Screenshots are stored privately and shown only to people with access to that run.
- An interrupted run is marked failed for review; it is not replayed automatically.
FAQ
Browser agent questions
Is this a real browser or a page scraper?
A real Chromium browser running in an isolated cloud sandbox. The agent observes page elements, navigates, records progress and can save screenshots, so it handles pages that need JavaScript.
Can the browser agent log in to sites?
For simple username-and-password logins, yes, using credentials stored in the encrypted workspace vault and bound to one exact site. The model never sees the secret, and authenticated pages are not screenshotted. Single sign-on, multi-factor codes and payment flows are not supported.
Will it fill in forms or buy things on its own?
No. Runs are read-only unless you approve a specific interactive run, password and payment inputs are refused, and scheduled agents cannot approve their own interactive work.
What does a browser run cost?
It is charged in credits from the measured sandbox compute and model tokens the run uses. The credit hold is reserved up front and the unused part is returned when the run settles.
Give the agent a page to check.
Connect what you already use, ask for the work in plain language, and approve what leaves the building.
7-day trial with a card on file and 300 usage credits. Then $49.99/month unless you cancel.