Stealth Browser Agent — Anti-Detect AI Browser Automation
Pricing
from $17.00 / 1,000 step completeds
Stealth web automation agent and AI browser agent for hard-to-automate sites: computer use and browser use on the web. Send a URL and a plain-language task; it navigates, fills and submits forms, runs multi-step flows, and returns structured JSON. Anti-detection fingerprinting, residential proxy.
Calling this again
reuse
Optional
The parameters for repeating this kind of task: the Actor to call, this run's input with URLs and the task text replaced by placeholders, and where current prices are listed. Useful for keeping notes on which tool handles which task. Contains no results.
Message
message
Optional
Optional feedback guidance when the task could not be completed or the answer was empty. The original error text is unchanged.
Final URL
url
Required
The URL of the page after all navigation and interactions completed. May differ from the starting URL if the agent followed links or was redirected.
Task
task
Required
Echo of the input task description, for traceability.
Status
status
Required
Outcome of the task. One of: completed (task finished successfully), failed (agent could not accomplish the task), timeout (hard deadline reached), max_steps_reached (step limit hit before completion), budget_exhausted (maxTotalChargeUsd reached before completion). Capped runs (max_steps_reached / budget_exhausted) may still carry a best-effort partial in result — check resultComplete.
Result
result
Required
Structured data extracted by the agent. Shape matches outputSchema if one was provided, otherwise best-effort JSON — usually an object, or a top-level array when the task asks for a bare list. Contains the information the task asked for. On a capped run this may be a best-effort PARTIAL synthesized from the data collected before the cap — fields without collected data are null, and resultComplete is false. Always check resultComplete before treating this as a full answer.
Result Complete
resultComplete
Optional
True only when status is completed and result has data. False means result is either empty, all null (see error), or a best-effort partial from a capped run — machine-checkable so agent consumers can decide to use the partial or re-run with a higher maxSteps / maxTotalChargeUsd.
Action Log
steps
Required
Ordered list of actions the agent took. Each entry records the action type, target, reasoning, and whether it succeeded. Useful for debugging and understanding agent behavior.
[ { "step": 1, "tool": "get_text", "args": { "selector": "h1" }, "result_summary": "Example Domain", "success": true }]Screenshot URL
screenshotUrl
Optional
Public URL of a full-page screenshot of the final page state, stored in the run's key-value store. Populated when a screenshot is captured.
Step Transcript URL
stepTranscriptUrl
Optional
Public URL of the capped full step transcript, stored in the run's key-value store. Populated on incomplete runs after at least one step.
Error
error
Optional
Human-readable explanation when status is not 'completed', or when a completed run's result has no data (every field null): why, and what to change in the task. Absent on successful runs with data.
Error class
errorClass
Optional
Machine-routable failure class. source_error is a temporary target-site failure and is retryable; source_walled is an access challenge and may be retryable; no_content, malformed_input, and agent_error are not retryable without changing the page, input, or task; not_covered means this Actor does not operate on that target, so re-running the same input will not change the answer; budget_exhausted is retryable with a larger budget. Null on success.
Retry advice
retry
Optional
Machine-readable retry guidance for this record, derived from errorClass. Null when the record is a result. 'never' = the source answered, or we do not cover this target: re-running the same input returns the same answer, do not re-run. 'with_changes' = the input for this record must change (see error for what): do not re-run unchanged. 'later' = a transient fault on the source or our side, or the run's budget/deadline ran out before this record: re-run the same input after a short delay.
Start URL (batch)
startUrl
Optional
In a batch run (urls), the start URL this row answers. Absent on single-URL runs.
Batch position
batchIndex
Optional
In a batch run (urls), this URL's position in urls, starting at 0. Absent on single-URL runs.
Evidence
evidence
Optional
Where each answered value in result appears on the pages the agent visited, checked automatically on every run (result itself is never changed). fields maps each string or number field's path (e.g. rooms[0].price) to a status: verbatim (appears exactly as given), normalized (appears reformatted: spacing, letter case, number separators or currency sign, or date format) or not_found (could not be matched on the captured pages; a number written with a single separator before exactly three digits, such as 1,234, is matched only when written the same way, since it can be read two ways). A match also gives step (the steps entry after which the page was captured; 0 = the starting page), pageUrl and a short excerpt of the page text around it; a value found in a page-content attribute (a link's or element's title, an image's alt text, an aria-label, a datetime, the meta description or og: tags) shows that attribute's line, e.g. "title attribute: …". Form fields, editable areas and hidden elements never count, so text the agent typed is not evidence. A null field is not_found with a blocker: human_verification, login_wall, run_ended (the run stopped first) or not_on_page. Booleans are not listed. The top-level blocker gives the run-wide reason, if any; truncated is true when not every field could be listed (more than 200 fields, or a very large run).