Glossary

AI Memory Engine

A system designed to store, organize, and retrieve information for AI applications, enabling them to maintain context and recall previous interactions or knowledge. AI memory engines help overcome the context limitations of large language models by providing persistent storage of information across multiple interactions.

DataPoints

The fundamental building blocks of cognee's knowledge graphs that represent structured information about entities and their relationships. DataPoints serve as the atomic units of knowledge, transforming unstructured data into a structured format that can be efficiently stored, queried, and analyzed within a knowledge graph.

Knowledge Graph

A structured representation of knowledge that uses graph structures with nodes (entities), edges (relationships), and properties to represent and organize information. Knowledge graphs transform messy, real-world data into organized networks that enable AI systems to understand context, resolve ambiguities, and infer hidden connections.

Semantic Search

A search method that understands the contextual meaning of search terms rather than just matching keywords, allowing for more relevant results based on intent and meaning. Semantic search captures the underlying intent of a query and returns results that are conceptually related, even when they don't contain the exact search terms.

Vector Database

A database optimized for storing and querying vector embeddings, which are numerical representations of data (like text or images) that capture semantic meaning. Vector databases convert data into multidimensional arrays of numbers that represent the essence and meaning of the content, enabling similarity searches based on concepts rather than exact matches.

Embedding

A numerical representation of data (such as text, images, or audio) in a vector space, where similar items are positioned closer together, enabling semantic operations. Embeddings capture the semantic essence of content, allowing AI systems to understand relationships between different pieces of information based on their meaning.

RAG

Retrieval-Augmented Generation (RAG) is an AI framework that enhances large language models by retrieving relevant information from external knowledge sources before generating responses, improving accuracy and reducing hallucinations. RAG combines the generative capabilities of LLMs with information retrieval systems to ground responses in factual data rather than relying solely on the model's internal parameters.

GraphRAG

GraphRAG combines the power of knowledge graphs with Retrieval-Augmented Generation, using graph structures to provide context-aware, relationship-rich information retrieval for more accurate and contextually relevant AI responses. This hybrid approach delivers both the contextual richness of vector searches and the structured accuracy of knowledge graphs for complex, multi-faceted queries.

Cognitive Architecture

A framework that describes how to organize the "mind" of an AI system—its way of processing, storing, and retrieving information. Cognitive architectures typically include components for memory, decision processes, and actions, inspired by human cognitive systems. These architectures help transform LLMs from simple text generators into sophisticated, human-like problem-solvers that can remember, learn, and adapt over time.

Working Memory

A component of AI cognitive architecture that acts as a short-term scratchpad holding immediate context, such as recent chat messages or partial solutions. Working memory enables an AI system to maintain awareness of the current conversation or task state. It's analogous to human short-term memory and helps AI systems maintain coherence during complex interactions.

Episodic Memory

A type of long-term memory in AI systems that stores records of past events and experiences. Episodic memory allows AI to recall "what happened" in previous interactions, enabling it to learn from past successes and failures. This memory type helps AI systems build a narrative understanding of their interactions and experiences over time.

Semantic Memory

A type of long-term memory in AI systems that contains factual knowledge about the world. Semantic memory stores general facts and concepts independent of personal experiences, providing AI with a foundation of knowledge to draw upon. Unlike episodic memory, semantic memory focuses on "knowing that" rather than "remembering when."

Procedural Memory

A type of long-term memory in AI systems that remembers how to perform tasks. Procedural memory might be embedded in the agent's code or the LLM's parameters, enabling the system to execute complex procedures without explicit instructions. This memory type focuses on "knowing how" to perform actions or solve problems.

Pipelines

Modular sequences of tasks in cognee that break down complex data processing into manageable steps. Pipelines connect tasks so that the output of one becomes the input for the next, creating a structured data flow. Pipelines enable efficient processing of information by handling data recursively, providing batching for improved performance, and scaling to accommodate intricate workflows.

Model Context Protocol (MCP)

An open standard introduced by Anthropic designed to connect AI systems with data repositories, business tools, and development environments. MCP standardizes how AI applications interact with various data sources, enabling them to generate more relevant, context-aware responses. MCP helps solve the problem of fragmented data by providing a common language for AI systems to communicate with different data platforms.

Hallucination

A phenomenon where large language models generate false or made-up information while presenting it as factual. Hallucinations occur partly because models don't have robust knowledge stores or mechanisms to verify the accuracy of their outputs. AI memory systems and retrieval-augmented approaches help reduce hallucinations by grounding responses in verified, external data sources.

Context Window

The limited amount of text a large language model can process at once during a single interaction. The context window determines how much previous conversation or document content the model can "see" when generating a response. When information exceeds the context window, the model "forgets" earlier content, highlighting the need for external memory systems for long-running interactions.