Browser actions

POST /v1/browse opens a page, executes actions sequentially, and extracts the resulting page with the same formats as POST /v1/scrape.

Supported actions

Wait for an element

{ "action": "wait", "selector": "main[data-ready=true]", "timeout_ms": 10000 }

Fill or type

Use fill for ordinary controlled inputs and type when the page depends on keyboard events.

{ "action": "fill", "selector": "input[name=email]", "value": "agent@example.com" }

{ "action": "type", "selector": "input[role=combobox]", "text": "Paris" }

Click

{ "action": "click", "selector": "button[type=submit]" }

Scroll

Offsets are CSS pixels and may be negative.

{ "action": "scroll", "x": 0, "y": 800 }

Sleep

Use a fixed delay only when the page has no observable selector to wait for.

{ "action": "sleep", "ms": 500 }

Result model

action_results is ordered exactly like the request. The final outputs block is compatible with the V1 extraction response.

{
  "action_results": [
    { "action": "fill", "ok": true, "detail": "ok" },
    { "action": "click", "ok": true },
    { "action": "wait", "ok": true }
  ],
  "outputs": { "markdown": "# Search results\n..." },
  "credits_used": 2,
  "billing": { "billable": true, "units": 2, "unit": "credit" }
}

Browse costs two credits per started browser minute. action_count remains in the response for observability but does not create a separate commercial quota.

Each request accepts at most 25 actions, 32 KiB per text value, 2,048 characters per selector, 5 seconds per sleep, and 15 seconds of cumulative sleep. Action requests bypass the render cache.

Practical examples

• Wait for JavaScript-rendered content