Glossary

Core data

Brain
An authenticated knowledge boundary used to scope every application read and write.

Resource
An ingested source such as text, a file, or structured data.

Passage / text chunk
A retrievable portion of source text. Search uses passages by default.

Entity / node
A stable graph object with identity, labels, description, and properties.

Relationship / predicate
A typed directed connection between two nodes.

Event hub
A first-class event node used when an action needs actor, target, time, or context. Static facts do not require one.

Observation
An optional model-generated note stored alongside other knowledge representations.

Retrieval

Context
Evidence assembled for an LLM, including passages and graph/history fields. It is not a ranked Search response.

Hit
One ranked Search result with an ID, score, optional component scores, snippet, labels, metadata, and catalog node mapping.

Channel
A Search candidate source such as passages, entities, events, communities, or plugin:<name>.

Candidate pool
The items retrieved before later filtering, reranking, personalization, and truncation.

Fusion
Combining rankings or normalized scores from multiple candidate sources.

Reranking
Reordering a bounded candidate head with a more expensive scoring function. It cannot recover missing candidates.

Facet
Counts derived from metadata on returned Search hits.

Personalization
A soft query-gated reordering signal applied to eligible catalog hits, not an authorization or hard filter.

Recommendation
A ranking around a target user or node rather than a free-text query.

Operations and extensions

Task
An asynchronous ingest work record scoped to the same brain as the request.

Plugin
Trusted Python code loaded into BrainAPI to add routes, tools, hooks, or ranking behavior.

MCP
Model Context Protocol, used to expose BrainAPI or documentation tools to compatible agent runtimes.

Shipped behavior
Implemented and test-backed behavior documented as part of the current contract.

Idea to test
A hypothesis for application evaluation, not a BrainAPI quality or latency claim.