Create a Page monitor
Use cURL and jq, and setPERPLEXITY_API_KEY to your API key. Add one target with type: "page" for each public URL you want to watch. Creating a monitor makes its first run due immediately and schedules later runs until you pause or delete it; see Monitor API pricing.
cURL
200 response is a monitor, not a completed run. Set MONITOR_ID to its id for the requests below. Use an idempotency key if your application retries creation.
This example reads results through the API without configuring a webhook. Use webhooks to receive change notifications without polling for results.
Understand the first run
The first successful fetch of each page saves its baseline and reports no change for that page. Retrieve the first run to check its status andtarget_errors; do not submit another /run request after creation.
cURL
queued, running, or an unknown status, keep polling with a bounded retry policy. For failed or skipped, inspect error. A succeeded run can still contain individual target_errors.
Set RUN_ID to data[0].id and retrieve that run:
cURL
Baseline run response
Baseline run response
Read page changes
Later runs compare fetched text with the saved version of each page. Detected differences are judged for meaningfulness before a kept change appears inchanges with type: "page_change" and a structured diff. The comparison reflects the fetched versions, not every intermediate edit.
Use the same run-history request to read later checks. A monitor watching release notes can return:
Page change response
Page change response
counts: {"candidates": 0, "changes": 0, "errors": 0}, changes: [], and target_errors: []. Changed text can also produce candidates that the relevance judge does not keep.
Page result fields
Fetch and processing errors are separate from changes. Each
target_errors entry contains type: "page", url, code, and message; a failed fetch preserves that page’s baseline. An error with the http_status code also includes the page’s status_code. Other codes include invalid_url, blocked, no_content, load_failed, timeout, and fetch_unavailable. Use the code for decisions rather than parsing the message, and allow for new codes.
A run can succeed when some pages fail and others are fetched successfully. If every source fails, the run is failed; inspect its error and target_errors.
Configure what to track
Choose pages
Supply 1 to 10 Page targets, each with its own URL. A Page monitor checks those URLs.
Duplicate page URLs are rejected after normalization of the scheme, host, default port, empty path, and fragment. Query strings remain significant for that validation, and the detector can deduplicate additional normalized equivalents when fetching.
To change the watched pages, replace the full
targets array. Include every URL you want to keep. Existing baselines are preserved. A Page URL with no stored baseline establishes one on its first successful fetch.
cURL
Set a relevance goal
Use the goal to describe which changes you want to keep, for example “New features and breaking changes in the release notes; ignore navigation and footer edits.” The goal filters detected text changes. Page judging evaluates whether the detected text change is meaningful. With a goal, it evaluates the change against those instructions; without one, it evaluates general meaningfulness. Successful runs advance the saved page version even when a detected change is not kept. Updating the goal therefore does not replay previously rejected changes.Next steps
Pause the monitor when you finish trying it. You can also change its schedule, add webhooks, or check the API reference for the full contract.Track API and pricing changes
Watch a changelog, pricing page, and API guide for changes that matter to you.