agenthub · knowledge.store
Knowledge Store
Write the embedded chunks into a knowledge namespace, replacing whatever those same pages stored last time.
Why the node earns its place
The end of the ingestion chain, and the reason re-indexing is safe: it names the source URLs in this batch as the ones being replaced, so running the same crawl again atomically swaps the stale chunks rather than doubling them.
How it works at run time
- 1Take the namespace from the wired port or the run.
- 2Write the chunks, declaring the batch's source URLs as replaced.
- 3Return how many were inserted, deleted and skipped as duplicates.
What comes out
inserted new chunks written deleted stale chunks replaced skipped_as_duplicate chunks already present, unchanged
What you wire
Inputs
- chunks *array
- namespace_idstring
Outputs
- insertednumber
- deletednumber
- skipped_as_duplicatenumber
Configuration
Nothing to configure — this node has no settings of its own.
What usually goes wrong
Watch for this
It writes as the buyer, not as the workflow's own runtime identity — the namespace belongs to the org, so anyone else's token is rejected. If every chunk write fails for a non-publisher, that is the cause rather than a broken namespace.
Knowledge Fetch reads what this writes. An agent that retrieves nothing usually means this step never ran, not that retrieval is broken.
Behaviour & provenance