AI browser agent

Give it a destination. It finds the path.

Describe a task in plain words. Pathfinder opens a real browser, reads the page, and clicks, types and scrolls until the job is done. You watch every step.

No card needed · Runs in isolated browsers · Full step replay
https://shop.example/games

Game Shop

Search games…
Search
Puzzle Quest
$4.99
Sky Racer
$2.99
Farm Days
$6.49
How it works

A loop, not a script.

No selectors to maintain and no recorded macros that break when a page changes. Every step, Pathfinder looks at the page fresh and decides what to do.

01 / LOOK

Reads the page

Takes a screenshot and numbers every button, link and text box it can use.

02 / DECIDE

Picks the next move

The AI compares the page with your goal and its notes, then chooses up to three actions.

03 / ACT

Does it for real

Clicks, types, selects, scrolls, switches tabs, and saves what it learns along the way.

↻ repeats until the task is done, then returns a written answer + recording
Use cases

Hand off the clicking.

Test your site

Walk through sign-up, checkout or every game on your site and report what breaks.

"Open each game on mysite.com and list any that fail to load"

Research

Visit several pages, compare what they say and bring back one clean summary.

"Compare the pricing pages of 3 game engines"

Admin chores

Fill in forms and update records in the web tools you already log into.

"Add these 5 products to my store dashboard"

Monitoring

Check a page through the API on your own schedule and get the result back as text.

"Is the 'Download' button on my landing page working?"
Built for trust

See everything it did.

Each run keeps a screenshot of every step, what the agent was thinking, and a replayable recording. Plug it into your own app with a simple API.

  • Live viewWatch the browser while it works and stop it any time.
  • Isolated browsersEvery task runs in a fresh browser with no shared cookies.
  • Step-by-step replayScreenshots and a GIF recording of the whole run.
  • You take the wheel when neededHit a CAPTCHA or a login code? It pauses and lets you click and type in the live view, then carries on. It never makes up an answer.
# start a task
curl -X POST https://your-domain/v1/tasks \
  -H "Authorization: Bearer pf_your_key" \
  -H "Content-Type: application/json" \
  -d '{"task":"Find the price of Sky Racer on shop.example","max_steps":15}'

# check on it
curl https://your-domain/v1/tasks/42 \
  -H "Authorization: Bearer pf_your_key"

# → {"task":{"status":"done","answer":"Sky Racer costs $2.99", ...}}
Pricing

Pay for steps, not seats.

Buy credits once and use them any time. 1 credit = 1¢. Each step (one look → decide → act) costs 1 credit on Gemini Flash. A simple lookup takes about 4–8 steps, so most tasks cost under 10¢.

Free

$0
  • 50 credits to try it
  • Live view + take control
  • All browsers and devices
Start free

Starter

$10
  • 1,000 credits
  • ≈ 125+ simple tasks
  • API access
  • Credits never expire
Get Starter

Pro

$49
  • 6,000 credits
  • Save 18%
  • Many browsers at once
  • All AI models + proxies
Get Pro

Business

$199
  • 30,000 credits
  • Save 34%
  • Bigger batches
  • Priority help
Get Business
Per stepCreditsCost
Gemini Flash (fast, default)11¢
GPT mini · Claude Haiku2–32–3¢
Gemini Pro · GPT · Claude Sonnet4–64–6¢
Residential proxy (add-on)+2+2¢

Prices in USD. You only pay for steps the agent actually takes. Exact prices are shown next to each option in your dashboard.

Questions

Good to know

What counts as a step?

Each time Pathfinder looks at the page and acts. A step costs 1 credit (1¢) on Gemini Flash, more on bigger models or with a proxy. Credits come off as the task runs, and you can set a maximum number of steps for every task.

Can it log into my accounts?

Only if you include the details in the task. Each task uses a fresh browser, so nothing carries over between runs.

What isn't allowed?

Spam, fake reviews, votes, clicks or followers, creating accounts in bulk, automated CAPTCHA solving, and anything against a site's rules. These tasks are refused automatically.

Which AI does it use?

Google Gemini, with the page screenshot and a numbered list of everything clickable. The model can be changed.

What if it gets stuck?

It notices repeats and tries another way. If a CAPTCHA, login or verification page blocks it, it hands the browser to you in the dashboard. If it still can't finish, it stops and explains why, so you only pay for the steps it took.

Is there an API?

Yes. Create a key in your dashboard and start tasks with one request. See the API docs.

Your next task could run itself.

Sign up free with 50 credits, type what you need, and watch Pathfinder do it.

Start free