Search available Actors (scrapers and data collectors) by keyword. Returns each Actor's id, title and description. Start here when you need to extract data from the web. Pass an empty search to list everything available.

Get the full JSON Schema for an Actor's input, plus its description. Call this before call-actor so you know which fields are required and how they are named.

Run an Actor and return its results. The shape of input depends on the Actor — read it from fetch-actor-details first. Returns a preview of the items plus a datasetId for retrieving the rest via get-dataset-items. A run that partially fails still returns what it collected; check get-actor-log for detail.

Get the status, item count and dataset id of a previous run. Use this after call-actor to check whether a run succeeded, failed, or was aborted.

List recent Actor runs in this server session, newest first.

Read the log of a run. This is where an Actor reports per-board or per-page failures that did not fail the run as a whole — check it when a run succeeded but returned fewer items than you expected.

Stop a run that is still in progress. Items already collected are kept.

Page through the results of a previous Actor run. Use the datasetId returned by call-actor. Use fields to project only the keys you need — this matters when items contain long text such as page content or job descriptions.

Get the item count of a dataset without reading the items themselves.

Infer the shape of a dataset from a sample of its items. Use this to learn which fields exist before requesting them via fields, instead of pulling whole items to find out.

Read a single record from a key-value store by key.

Save a named Actor configuration so a search worth repeating gets an id instead of a body of JSON to retype. Re-using an existing id updates it.

List saved tasks, or fetch one by id.

Remove a saved task. The runs and datasets it produced are kept.

Run a saved task. Any fields given in input override the saved ones for this run only, so one task covers a family of similar searches.

Re-run a finished run with the same Actor and input, optionally with a longer timeout — the usual reason a run needs resurrecting. Creates a new run, so the original's results and diagnosis stay intact.

Run an Actor or a saved task on a cron schedule, executed by the server itself. Give either actor with inline input, or task. Re-using an id updates it.

List schedules with their next and last run times.

Remove a schedule. Runs it already produced are kept.

Delete datasets and stores older than keepDays. Every run creates a dataset, so these accumulate; configuration stores are never removed. Use dryRun first.

Record that you have applied to a role or dismissed it. Marked roles stop appearing in future digests, which is what turns the digest from a feed into a working list. Accepts several URLs at once.

List roles recorded as applied or ignored, newest first.

Lists roles applied to that have gone quiet, longest wait first. Anything that has had a response — including a rejection — is excluded, and an application already chased inside the window is not chased again.

Note that an employer replied, rejected, invited you to interview or made an offer. Stops that application appearing in follow-ups.

Records that you followed up, so the same silence is not flagged again immediately.

Forget that a role was applied to or ignored, so it can appear again.