vocabulary:
  title: Cognee API Vocabulary
  description: Canonical terms and definitions used across the Cognee AI memory and knowledge graph platform
  version: "1.0.0"
  source: https://docs.cognee.ai/api-reference/introduction

terms:

  - term: dataset
    definition: >
      A named container for ingested data within Cognee. Each dataset holds one or more
      data items (files, URLs, documents) and its associated knowledge graph. Datasets
      are owned by a user and subject to permission-based access control.
    type: resource
    relatedEndpoints:
      - GET /api/v1/datasets
      - POST /api/v1/datasets
      - DELETE /api/v1/datasets/{dataset_id}

  - term: cognify
    definition: >
      The core intelligence pipeline that transforms raw ingested data into a structured
      knowledge graph. The six-stage ECL process includes: document classification,
      text chunking, entity extraction via LLM, relationship detection, vector embedding
      generation, and content summarization. Also refers to the POST /api/v1/cognify endpoint.
    type: operation
    acronym: ECL (Extract, Cognify, Load)
    relatedEndpoints:
      - POST /api/v1/cognify

  - term: knowledge_graph
    definition: >
      A graph-structured representation of entities and their relationships extracted
      from ingested data. Stored in a hybrid graph-vector-relational backend. The graph
      is the primary data structure queried by search endpoints.
    type: data-model
    aliases:
      - graph

  - term: search_type
    definition: >
      An enumerated value specifying which retrieval strategy to use when querying the
      knowledge graph. Cognee supports 16 search types including semantic, graph-based,
      RAG, temporal, agentic, and Cypher query modes.
    type: enumeration
    values:
      - SUMMARIES: Return summarized content nodes
      - CHUNKS: Return raw text chunks
      - RAG_COMPLETION: Retrieval-augmented generation using vector similarity
      - TRIPLET_COMPLETION: Complete answers using graph triplets
      - GRAPH_COMPLETION: LLM answer synthesis using graph traversal (default)
      - GRAPH_COMPLETION_DECOMPOSITION: Decompose query before graph completion
      - GRAPH_SUMMARY_COMPLETION: Graph completion over summarized nodes
      - CYPHER: Execute a Cypher graph query directly
      - NATURAL_LANGUAGE: Natural language to graph query translation
      - GRAPH_COMPLETION_COT: Chain-of-thought graph completion
      - GRAPH_COMPLETION_CONTEXT_EXTENSION: Graph completion with extended context window
      - FEELING_LUCKY: Best-effort automatic search mode selection
      - TEMPORAL: Search for time-anchored entities and events
      - CODING_RULES: Search for coding patterns and rules
      - CHUNKS_LEXICAL: Lexical (keyword) chunk retrieval
      - AGENTIC_COMPLETION: Multi-step agentic search with tool use

  - term: node_set
    definition: >
      A list of string labels used during data ingestion to tag and group related data
      points within the knowledge graph. Node sets enable targeted search scoping via
      the node_name parameter on search requests.
    type: concept
    relatedFields:
      - add.node_set
      - search.node_name

  - term: pipeline_run
    definition: >
      A single execution of the add or cognify pipeline against one or more datasets.
      Each pipeline run has a unique UUID and a lifecycle status of pending, running,
      completed, or failed. Background runs can be monitored via the WebSocket endpoint.
    type: resource
    statuses:
      - pending
      - running
      - completed
      - failed

  - term: agent
    definition: >
      An AI agent identity in Cognee, represented as a system user with a dedicated
      API key. Agents can interact with the Cognee API autonomously using their own
      credentials, enabling multi-agent memory architectures.
    type: resource
    relatedEndpoints:
      - GET /api/v1/agents/list
      - POST /api/v1/agents/create

  - term: ontology
    definition: >
      A structured vocabulary or taxonomy (in RDF/OWL format) that can be uploaded to
      Cognee and referenced during the cognify pipeline to constrain and guide entity
      extraction and knowledge graph construction.
    type: concept
    relatedFields:
      - cognify.ontology_key

  - term: add
    definition: >
      The ingestion operation that accepts files, HTTP URLs, or GitHub repository URLs
      and stores them in a dataset for subsequent cognify processing. Corresponds to
      the POST /api/v1/add endpoint.
    type: operation
    relatedEndpoints:
      - POST /api/v1/add

  - term: chunk
    definition: >
      A segment of text produced by splitting an ingested document during the cognify
      pipeline. Chunk size is configurable via the chunk_size parameter. Chunks are
      the atomic unit for LLM processing and vector embedding.
    type: data-model
    relatedFields:
      - cognify.chunk_size
      - cognify.chunks_per_batch

  - term: vector_store
    definition: >
      The vector database backend used by Cognee to store embeddings for semantic
      similarity search. Supported providers include LanceDB, ChromaDB, and pgvector.
    type: infrastructure
    supportedProviders:
      - lancedb
      - chromadb
      - pgvector

  - term: llm_provider
    definition: >
      The Large Language Model service used by Cognee for entity extraction, graph
      construction, and completion-type searches. Supported providers include OpenAI,
      Anthropic, Google Gemini, Mistral, and Ollama (local).
    type: infrastructure
    supportedProviders:
      - openai
      - anthropic
      - gemini
      - mistral
      - ollama