Knowledge bases
Managing and inspecting knowledge sources
The knowledge-base view shows each source’s progress, indexed text, usage, and last refresh. Members can inspect shared sources; the member who owns a knowledge base can change its sources, cancel ingestion, or delete it.
- Availability
- All plans
- Configured in
- Stand app → Stand-ins → Knowledge bases → Sources, or stand knowledge commands
- Category
- Knowledge bases
- Reference status
- Current
Before you begin
Plan availability: All plans for source types and inspection surfaces available to the current plan.
Prerequisite: Organization membership to view shared knowledge; ownership of the knowledge base to change its sources.
Key boundary: Inspection shows extracted and indexed text, not a pixel-perfect copy of the original page or document and not a simulation of every passage the model might retrieve for every question.
Knowledge hierarchy
| Level | What it represents |
|---|---|
| Knowledge base | An organization-visible container with a description, owning rep, sources, and Stand-in attachments. |
| Source or document | One website crawl, sitemap, uploaded document, or replacement lifecycle inside the knowledge base. |
| Task | The active or completed ingestion, refresh, replacement, or cancellation work for that source. |
| Indexed section | A searchable unit of extracted text retained with source title, URL or filename, and section context. |
What can be inspected
- Source type, URL or filename, task and source status, last refresh, and customer-visible usage.
- Status badges use a static clock for queued work, an activity indicator while ingestion is running, and static semantic icons for completed, failed, or cancelled work.
- Successfully indexed pages or documents and their extracted text.
- Indexed sections with retained structural context.
- Search within a document’s indexed sections to confirm that an expected phrase or topic was extracted.
- Failure and active-ingestion state needed to decide whether to refresh, upload a revised document, cancel, or preserve the current source.
Owner actions
| Action | Effect |
|---|---|
| Re-crawl, Re-ingest or CLI upload replacement | Refreshes a website, sitemap or uploaded source while retaining prior successful content until success. The Stand app updates files through a separate new-upload-and-delete workflow. |
| Cancel ingestion | Stops supported active work without promoting incomplete content. |
| Delete source or document | Removes the selected source and its derived content, or the selected document and its indexed sections. Source deletion also cancels its open tasks. |
| Delete knowledge base | Removes the owned container and its content and cancels its open tasks. Attached Stand-ins must be inspected; CLI deletion requires explicit --force-detach when any are attached. |
CLI knowledge management
- stand knowledge list returns shared bases, ownership, capabilities, usage and ingestion availability. create and update manage the owned base’s name and description.
- sources, documents and tasks list ingestion resources. documents get reads extracted text, documents chunks reads indexed sections, and documents search searches one document. knowledge search searches the base; both searches accept top-k from 1 to 20, defaulting to 5.
- upload adds or replaces an approved document; crawl and sitemap queue supported website ingestion. sources recrawl, sources refresh and tasks cancel manage refresh and unfinished work.
- knowledge agents lists attached Stand-ins. Before knowledge delete, the CLI previews connected Stand-ins and requires confirmation plus --force-detach if attachments exist. Force detach removes the base from all connected Stand-ins, including other members’ Stand-ins and connections added after the preview; the preview does not lock the attachment list.
- Reads require knowledge:read; owner mutations require knowledge:write. Named base deletion also requires knowledge:read for its attachment preview. --yes skips the CLI prompt only for a mutation the user has already approved.
- Ordinary document, chunk and search reads expose source text. The API’s bounded inspection mode returns metadata without document text or raw ingestion errors.
Visibility, ownership, and downgrade state
- Knowledge bases are visible across the organization but retain one owning rep.
- Non-owning members can inspect shared knowledge and select available bases for their Stand-ins within plan limits.
- Only the owner can change or delete the knowledge base and its sources.
- Member removal can transfer knowledge ownership so shared resources are not orphaned.
- After downgrade, over-limit knowledge remains visible in grace or read-only state; ingestion and refresh pause until usage fits the current entitlement.
Interpretation boundaries
- Successful extraction proves that text was indexed, not that the source is accurate or current.
- Document and whole-base search use the same ranking method over different scopes. Scores are relative relevance signals, can exceed 1, and are not confidence percentages or comparable across queries or scopes. Preserve the returned order and inspect the text rather than sorting by score or applying a universal cutoff.
- Deleting a source changes future retrieval and does not rewrite messages already generated from earlier knowledge.
- The visitor never receives the private inspection interface or raw indexed sections as hidden system messages.
Example
The screen below shows the feature in its normal Stand context. Labels and surrounding controls may vary with account state and plan.
