PendiumDocs

Search the knowledge base

Semantically search a brand's knowledge base, list its sources, and remove standalone sources.

A brand's knowledge base holds its website content, uploaded documents, and training materials. Reach for these tools to find specific facts, claims, or positioning language before writing content, to see what has been ingested, or to prune a source that's no longer wanted.

When to use

Use query_knowledge to pull the exact language or facts you need mid-writing. Use list_knowledge_sources to audit what's ingested, and delete_knowledge_source to remove a standalone source. For a broader company overview before diving into specifics, combine with get_factsheet.

Typical sequence

  1. query_knowledge — semantic search for the fact or phrasing you need.
  2. list_knowledge_sources — see everything that's ingested (to audit or find a source id).
  3. delete_knowledge_source — remove a standalone source by id.

Tools

query_knowledge

Semantic search over a brand's knowledge base. Returns the most relevant text chunks ranked by similarity.

Authentication: Required.

ParameterTypeRequiredDefaultDescription
syntheticIdintegerYesThe agent ID to search knowledge for.
querystringYesNatural language search query, e.g. "pricing tiers and features", "company founding story".
topKinteger (1–20)No8Number of results to return.

Each chunk carries a score (0–1 similarity, higher is more relevant), text, sourceType ("webpage", "document", "training"), and sourceUrl when applicable. Use natural-language questions rather than keywords, and raise topK when you need more context.

list_knowledge_sources

List all knowledge sources for a brand — website pages, uploaded documents, and training materials — showing what content has been ingested.

Authentication: Required.

ParameterTypeRequiredDescription
syntheticIdintegerYesThe agent ID to list knowledge sources for.

Returns sourceCount plus a sources[] array; each source has an id, type ("webpage", "document", "training"), origin (URL or filename), status ("completed", "processing", "failed"), contentPreview (first ~200 chars), and createdAt.

Not listed here: analyzed-URL sources (managed via the analyzed-URL tools) and social profiles (LinkedIn company / YouTube channel, saved via the Profile tab or save_social_handles and managed by editing the brand's social handle). This matches the web app's Knowledge list.

delete_knowledge_source

Permanently remove a knowledge source and all of its indexed vectors. Destructive and irreversible — anyone with access to the agent, including shared team members, can delete one. Pass only the sourceId; the underlying vector ids are resolved server-side. Use list_knowledge_sources to find the id.

Authentication: Required.

ParameterTypeRequiredDescription
syntheticIdintegerYesThe agent the knowledge source belongs to.
sourceIdintegerYesThe knowledge-source id to delete (the id from list_knowledge_sources).

Returns deletedSourceId and deletedVectorCount. This tool cannot delete an analyzed-URL source (manage via delete_analyzed_url) or a social-profile source (manage by editing the brand's social handle) — both are hidden from list_knowledge_sources, and passing such an id returns an error explaining where to manage it.

On this page