Knowledge
AI knowledge base (RAG) — training material chunked and embedded for retrieval.
List knowledge sources
A page of the tenant's AI knowledge sources, newest first. Admin only.
Create a knowledge source
Create and ingest a knowledge source (admin only).
Get knowledge-base usage
Current chunk count vs the tenant's plan ceiling (admin only). Drives
Can the workspace use a knowledge base?
Whether a usable embedding model + credential exists (doc 103 §I). Not
Knowledge health snapshots
The latest weekly knowledge-base quality measurements, newest first
List background knowledge operations
The knowledge base's long-running jobs, newest first — learning from
Start a background knowledge operation
Queues an operation (doc 103 §D). `learn_from_history` needs AI
Dry run of a background knowledge operation
What the operation would do, without doing it: e.g. for `dedupe` the
Get a background knowledge operation
One operation's status (`queued`, `running`, `paused`, `cancelled`,
Tail an operation's process log
Log lines oldest first; pass `after` (the last line's id) to fetch only
Pause an operation (it stops after its current step)
Pause an operation (it stops after its current step)
Resume a paused operation from where it stopped
Resume a paused operation from where it stopped
Cancel an operation
Releases what it holds (a staged generation is dropped; live knowledge is untouched).
Knowledge-base evaluation summary
The 'Knowledge health' card's evaluation block (doc 103 §6, Phase 1):
List the gold-set questions
Oldest first. `last` is how the newest evaluation run did on the
Add a question to the gold set
An operator's own question (`origin = manual`, kept through every
Reword a gold-set question
The question becomes `manual` and survives regenerations. Requires knowledge management.
Remove a question from the gold set
Requires knowledge management.
List evaluation runs
Newest first. Requires knowledge or AI management.
Get one evaluation run
Get one evaluation run
Compare two evaluation runs
Usually the run before a retrieval change (`base`) and the one after
Change the evaluation settings
`auto_eval` (doc 103 §12, default on): when the weekly knowledge
Refresh a web source now (stage, then swap)
Queues a `refresh_source` operation: the source is re-read into a new
Get a knowledge source
One source's metadata, processing status and how many chunks it was
Update a knowledge source
Edit a knowledge source's title and training material (admin only).
Delete a knowledge source
Removes the source and all of its embedded chunks (admin only).
List a curated site source's discovered / ingested pages
Curated crawl (SC-1): the discovered link list for the review UI. Paginated; filterable by `search` (url substring) and `status`.
Queue selected discovered pages for ingest
Curated crawl (SC-1): mark discovered pages `pending` and start the batched crawl. `all: true` selects every discovered page (optionally narrowed by `search` — 'select all filtered').
Remove one page from the training data
Curated crawl (SC-1): delete a page's chunks and drop its frontier row so a resume won't re-ingest it.