What the browser agent does for you
Plenty of useful information never reaches an API. Competitor pricing pages, supplier portals, job boards, review sites and government registers are built for people with a browser. A Crons browser action gives your scheduled agent that same browser. You describe the job in one sentence, for example "Every weekday at 7 AM open the pricing pages of our 12 competitors and post any plan or price change to #pricing in Slack", and Crons turns it into a cron with fetch, browser and reasoning steps.
Each run the agent loads the page, waits for it to render, expands hidden sections when needed, reads the content and hands it to an AI reasoning step. That step pulls out the fields you care about, such as plan names, prices, stock status or a new job title, and the memory step compares them with the previous run. If nothing changed, nothing is sent. If something changed, you get one clear message with the old value and the new value.
- Open pages that need JavaScript to render, including single page apps
- Click tabs, accept cookie notices, page through results and expand accordions
- Fill search forms and filters, then read the result list
- Log in to portals with credentials you store in Crons and never paste into a prompt
- Capture a screenshot of the page as part of the run log
When to use a browser action instead of a web fetch
Crons has two ways to read the web. A web fetch downloads the page as served and is fast and cheap. A browser action runs a full headless browser and is counted in browser minutes on your plan. The planner picks the lighter option when it can, and you can switch any step by hand in the editor.
| Situation | Web fetch | Browser action |
|---|---|---|
| Static page or blog post | Best choice | Works but uses browser minutes |
| Prices loaded by JavaScript after page load | Often misses the data | Best choice |
| Content behind a login form | Not supported | Best choice with stored credentials |
| Search form with filters | Only if the result has its own URL | Best choice |
| RSS feed or JSON endpoint | Best choice | Not needed |
| Need a screenshot as proof | Not available | Included in the run log |
A practical rule is to try a web fetch first in a dry run. If the output in the run log is missing the numbers you expect, switch that step to a browser action and run the dry run again.
How a scheduled browser run works
- 1The schedule fires at the time and timezone parsed from your sentence, or you press Run now.
- 2The agent opens each target page in an isolated headless browser session.
- 3It performs the clicks, form inputs and waits defined in the step.
- 4An AI reasoning step extracts structured fields from what the page shows.
- 5The memory step compares the result with the last run and keeps only new or changed items.
- 6Delivery steps post the summary to Slack, add rows to Google Sheets or send an email through Gmail.
If a page times out or returns an error, the step retries up to 3 times with backoff. If it still fails, the run is marked failed, every attempt is visible in the log and you get an alert by email or Slack. A failed page never produces a fake "no change" result.
Browser minutes by plan
Browser minutes measure the time a headless browser is open for your crons. A simple page visit usually takes well under a minute. Usage counts per billing period, deleting a cron never returns minutes, and you get an alert at 80 percent before a hard stop at the limit.
| Plan | Browser minutes per month | Minimum frequency | Price yearly, per month |
|---|---|---|---|
| Starter | 60 | Every 1 hour | $49 |
| Pro | 300 | Every 15 minutes | $149 |
| Teams | 1,200 | Every 5 minutes | $499 |
As a hypothetical example, checking 12 competitor pricing pages once every weekday at about 20 seconds per page uses roughly 4 minutes a day, or around 85 minutes a month. That fits the Pro plan with plenty of room. Full details sit on the Crons pricing page.
Logins and credentials stay out of prompts
When a browser step needs to sign in, you save the username and password, or a session token, in the credential store of your workspace. The step references the credential by name. The value is encrypted at rest, never shown in the run log and never sent to the AI model provider as part of a prompt. You can rotate or delete a credential at any time and every cron that uses it picks up the change on the next run.
Browser actions only read and click where you tell them to. Steps that could have an outside effect, such as submitting a form that sends a message or places an order, can sit behind an approval gate, so a person confirms before the click happens. Payments are never executed by a cron. Read more on the security page.
Jobs teams give the browser agent
- Competitive intelligence that tracks plan names, prices and feature lists on rival pricing pages
- Recruiting teams that read new postings on niche job boards and add them to a sheet
- Finance teams that download the latest statement summary from a supplier portal
- Marketing teams that check whether partner pages still link to them
- Customer success teams that read status pages of tools their customers rely on
Start from the competitor monitor template and change the list of URLs, or write your own sentence on the homepage and see the parsed steps before you create an account.
Frequently asked questions
What is an AI browser agent in Crons?
It is a scheduled cron that uses a headless browser step to open pages, click, fill forms and read content, followed by an AI step that extracts the data you asked for and a memory step that keeps only what changed since the last run.
Can the browser agent log in to websites?
Yes. You store the login in the encrypted credential store and reference it in the step. The value never appears in prompts or run logs, and you can revoke it at any time.
How are browser minutes counted?
Minutes count the time a headless browser session is open for your crons in the current billing period. Starter includes 60, Pro 300 and Teams 1,200 per month. Deleting a cron does not return minutes.
What happens when a website changes its layout?
The AI step reads the page by meaning rather than fixed positions, so small layout changes usually keep working. If the expected data disappears, the run is flagged in the log and you get an alert instead of a silent empty result.
Will the agent submit forms or buy things on its own?
Only if you build that step, and you can put any such step behind an approval gate. Payments are never executed by a cron.
Do I need a browser action for every website?
No. Static pages and feeds work with a cheaper web fetch step. Use a browser action when content loads with JavaScript, sits behind a login or needs clicks to appear.