RAG Delta Sync
Pricing
$20.00 / 1,000 completed delta reports
Compare crawler exports and prepare changed chunks with stable IDs for a RAG index. Keep unchanged text out of the embedding queue. Ready to refresh your knowledge base?
Dataset ID containing the current crawl export. Rows need url and markdown, text or pageContent. Choose this OR currentPages.
Previous SOURCE crawl export using the same chunk settings. Do not point this at a delta-report output dataset.
Alternative to currentDatasetId: objects with url and text or markdown. No website is fetched by this Actor.
Alternative to previousDatasetId. Omit for an initial import.
Default false blocks all deletes. Enable only for complete crawl exports with complete page text. Coverage, source errors and delete ratio gates still apply.
Fraction of previous URLs that must remain valid in the new crawl before any delete is approved. New unrelated pages cannot hide lost coverage.
Blocks deletions if the proposed deleted chunks exceed this fraction of previous chunks. Review large rewrites before increasing.
Fails before a partial comparison if either source contains more rows. Maximum 200.
Paragraph-aware target size, not an exact token count. Keep this setting unchanged between comparisons.
Maximum UTF-16 code units per source page. Oversized pages fail; no silent truncation.
Hard cap on either snapshot chunk count and the combined upsert/deletion candidate plan. Exceeding it fails without a partial report.
