Knowledge operators

The Knowledge family contains 4 operators. Vector-backed semantic search over per-document collections.

Operators in this family

• knowledge.delete
• knowledge.list
• knowledge.search
• knowledge.upsert

knowledge.delete

Removes chunks from a knowledge namespace, optionally scoped to a single doc_key.

Version 1.0.0 · Semantics side_effect · Idempotent · Visibility advanced

Config

Field	Type	Required	Default	Description
doc_key	string	no	—	Doc identifier to delete. Omit to delete the entire namespace.
namespace	string	yes	—	Knowledge base name.

Input ports

Port	Type	Required	Description
input	any	no	Input

Output ports

Port	Type	Required	Description
output	object	no	Output

knowledge.list

Paginates chunk metadata for a knowledge namespace.

Version 1.0.0 · Semantics passthrough · Idempotent · Visibility advanced

Config

Field	Type	Required	Default	Description
limit	integer	no	50
namespace	string	yes	—	Knowledge base name.
offset	integer	no	0

Input ports

Port	Type	Required	Description
input	any	no	Input

Output ports

Port	Type	Required	Description
output	object	no	Output

knowledge.search

Semantic search over a pgvector knowledge namespace via cosine similarity.

Version 1.0.0 · Semantics passthrough · Idempotent · Secrets OPENROUTER_API_KEY

Config

Field	Type	Required	Default	Description
embedding_model	string	no	"openai/text-embedding-3-small"	OpenRouter-routed embedding model. Must match the upsert model.
filter	object	no	—	JSON-containment filter applied to chunk metadata.
namespace	string	yes	—	Knowledge base to search.
query	string	no	—	Query text. Falls back to the input port string if omitted.
top_k	integer	no	5

Input ports

Port	Type	Required	Description
query	string	no	Query

Output ports

Port	Type	Required	Description
output	object	no	Output
results	array	no	Results
scores	array	no	Scores

knowledge.upsert

Chunks, embeds, and persists text into a pgvector-backed knowledge base.

Version 1.0.0 · Semantics side_effect · Idempotent · Secrets OPENROUTER_API_KEY

Config

Field	Type	Required	Default	Description
chunk_size	integer	no	1500	Maximum characters per chunk.
content	string	no	—	Text to chunk and embed. Falls back to the input port string if omitted.
doc_key	string	yes	—	Stable identifier for the source document; re-upserts replace existing chunks.
embedding_model	string	no	"openai/text-embedding-3-small"	OpenRouter-routed embedding model. Must produce 1536-dim vectors.
metadata	object	no	{}	Free-form JSON metadata attached to every chunk for later filtering.
namespace	string	yes	—	Logical knowledge base name (e.g. ‘product-docs’).

Input ports

Port	Type	Required	Description
content	string	no	Content

Output ports

Port	Type	Required	Description
output	object	no	Output

Source schemas

Every operator above is rendered from its JSON schema in /opt/glyph/schemas/operators/:

• schemas/operators/knowledge.delete.json
• schemas/operators/knowledge.list.json
• schemas/operators/knowledge.search.json
• schemas/operators/knowledge.upsert.json