Refresh a source
One verb for every kind: re-fetches the page (url), re-discovers and re-fetches the site’s pages (website), re-processes the stored content (text/file). Asynchronous — 202 with the source in updating; poll the source until ready, attention or failed. Every refresh writes an update run (see /updates).
Authorizations
Your secret key (pep_sk_…) from Settings → API. Keys are org-scoped; treat them like passwords.
Path Parameters
Response
Accepted; the refresh runs in the background.
"knowledge_source"text, url, file, website Aggregated from the source's entries: ready (everything ingested), updating (work in flight — poll), attention (usable with some failed entries), failed (nothing usable).
ready, updating, attention, failed Slugs of the agents this source is active for.
Entry counts by status (one entry per page for websites).
The page (url) or site root (website).
Single-entry kinds: why processing failed, human-readable. Websites report per-entry reasons under /entries.
What the agent understood from this source (generated after each refresh).
website kind only.
When the last refresh run landed — the freshness stamp.
Full text. Present on retrieve/create/update for text/url/file; websites read per-page content via /entries.

