Skip to main content
POST
A source is indexed once, when you add it. Nothing re-crawls or re-reads it on a schedule. If a web page is edited or a document is revised after it was added, the agent keeps answering from the version that was indexed until you refresh it.

​
Which operation do I need?

Re-indexing a file source re-reads the copy already stored for it, which does not change. It cannot pick up edits you made to the document on your own systems — use Replace a Source File for that.

​
Re-index a Knowledge Base

POST /v1/knowledge-bases/{knowledgeBaseId}/reingest Queues a rebuild for every source in the knowledge base in one call. Takes no request body. URL sources are re-fetched from their live page; file sources are re-processed from their stored copy. Sources that already have a recent queued or running job are left alone and listed in skipped, so calling this twice in a row is safe. A source keeps its existing vectors through the slow part of the run — download, extraction, chunking and embedding — and swaps them only at the end, so the window in which it is not searchable is brief rather than instant. Other sources in the knowledge base are unaffected, since the worker processes one source at a time.
​
string
required
Knowledge base ID (UUID).
A knowledge base with no sources returns sourceCount: 0 and empty arrays. A knowledge base with more than 100 sources returns 400 with code too_many_sources; re-ingest those sources individually instead. The limit exists because ingestion is processed one job at a time. Nothing is queued when this happens — the call fails rather than silently refreshing the first 100.
400

​
Following progress

Poll each returned jobId with Job Status, or poll List Sources and wait for every source to leave pending. A finished source returns to active with a fresh last_synced_at; one that failed becomes error with the reason on its job.
A refresh that extracts no text is reported as a failure, and the old content keeps serving. The job ends failed with the reason in error_text, and the source becomes error — it does not return to active.The previously indexed content is deliberately left in place, so the agent keeps answering from the older version rather than losing the knowledge base to a temporary fetch problem. Because nothing was re-indexed, chunk_count and last_synced_at keep their previous values and continue to describe the content that is actually being used.Common causes: a URL that now returns 404 or blocks crawlers, a page whose content could not be summarised, or a replacement document that is a scanned image with no text layer. Fix the source and refresh again — an error source stays eligible for future refreshes.

​
Replace a Source File

POST /v1/knowledge-sources/{sourceId}/replace Uploads a new version of an existing file source’s document, overwriting the stored copy. The source keeps its id, so anything referencing it keeps working, and no outdated duplicate is left indexed alongside the new version. Only file sources can be replaced. A URL source returns 400 — re-ingest it instead, which re-fetches the live page.
​
string
required
Name of the replacement file. Becomes the source’s new display name.
​
string
required
MIME type of the replacement file. May differ from the original — you can replace a PDF with a DOCX. Must be one of the supported types: application/pdf, the DOCX type, text/plain, text/markdown, text/html or text/csv. Anything else returns 400.
​
integer
required
Size in bytes. Maximum 50 MB.
​
string
Optional hash of the new contents, stored for later duplicate detection.
Three steps, mirroring file source creation:
1

Request the upload

POST here to get a presignedUrl and a jobId. This immediately renames the source to fileName and marks it pending, so complete all three steps — stopping after this one leaves the source described by metadata that does not match its stored content.
2

Upload the file

PUT the file bytes to presignedUrl, sending a Content-Type header matching fileMime — the signature is bound to it and S3 rejects a mismatch. The URL is valid for 5 minutes. Until this succeeds the previous bytes stay in place and searchable.
3

Rebuild the vectors

POST /v1/knowledge-sources/{sourceId}/enqueue with that jobId and forceReindex: true.

​
In the Dashboard

The same two operations are on the knowledge base page:
  • Refresh all above the source list re-indexes every source.
  • The upload control on a file source’s row replaces its document with an updated one.
Refresh all reports how many sources it queued and shows each row’s Status as it works. Re-ingesting or replacing a single source additionally shows that job’s stage and chunk counts.