> ## Documentation Index
> Fetch the complete documentation index at: https://docs.cognee.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Llms core

# Cognee Core Documentation

> Foundational docs for getting started, core concepts, setup, guides, examples, and contributor-facing documentation.

See the full overview at [https://docs.cognee.ai/llms.txt](https://docs.cognee.ai/llms.txt).

## Getting Started

* [Introduction](https://docs.cognee.ai/getting-started/introduction.md): Cognee organizes documents, databases, and conversations into AI memory using knowledge graphs and vector search for accurate retrieval at scale.
* [Installation](https://docs.cognee.ai/getting-started/installation.md): Install Cognee on Python 3.10 to 3.14, set up a virtual environment, and configure an LLM provider before running your first remember and recall call.
* [LLM Quickstart Skill](https://docs.cognee.ai/getting-started/llm-quickstart-skill.md): Copy a Claude-ready skill that walks an LLM assistant through installing Cognee, picking providers, configuring keys, and running a first smoke test.
* [Quickstart](https://docs.cognee.ai/getting-started/quickstart.md): Run your first Cognee example after installation - call remember to ingest text, then recall to query the resulting graph with graph completion.

## Core Concepts

* [Core Concepts Overview](https://docs.cognee.ai/core-concepts/overview.md): Introduction to Cognee's core concepts, dual storage architecture, main operations like remember and recall, plus modular task and pipeline design.
* [Architecture](https://docs.cognee.ai/core-concepts/architecture.md): How Cognee combines relational, vector, and graph stores into a unified architecture for searchable, connected memory across components.
* [DataPoints](https://docs.cognee.ai/core-concepts/building-blocks/datapoints.md): DataPoints are Pydantic-based atomic units of knowledge in Cognee, carrying content, provenance, and indexing hints across vector and graph stores.
* [Tasks](https://docs.cognee.ai/core-concepts/building-blocks/tasks.md): Tasks are Cognee's smallest executable units, wrapping any Python callable with batching, error handling, and stream-based composition for pipelines.
* [Pipelines](https://docs.cognee.ai/core-concepts/building-blocks/pipelines.md): Pipelines orchestrate ordered Tasks into reproducible workflows with layered execution, per-dataset processing, status tracking, and access checks.
* [PipelineContext](https://docs.cognee.ai/core-concepts/building-blocks/pipeline-context.md): PipelineContext is a typed dataclass auto-injected into task functions, carrying user, dataset, data item, and extras through a pipeline run.
* [Remember](https://docs.cognee.ai/core-concepts/main-operations/remember.md): The remember operation is the main Cognee v1.0 ingestion entry point, writing to permanent graph memory or to per-session cache in a single call.
* [Recall](https://docs.cognee.ai/core-concepts/main-operations/recall.md): The recall operation is the main Cognee v1.0 retrieval entry point, auto-routing queries across session cache and graph-backed permanent memory.
* [Improve](https://docs.cognee.ai/core-concepts/main-operations/improve.md): The improve operation enriches an existing Cognee graph, bridges session memory into permanent memory, applies feedback, and builds global context indexes.
* [Forget](https://docs.cognee.ai/core-concepts/main-operations/forget.md): The forget operation is the unified Cognee v1.0 deletion command for removing data items, datasets, memory-only state, or all user-scoped content.
* [Add](https://docs.cognee.ai/core-concepts/main-operations/legacy-operations/add.md): Legacy ingestion operation that normalizes files, directories, or raw text into a dataset, preparing them for cognify to build vectors and graphs.
* [Cognify](https://docs.cognee.ai/core-concepts/main-operations/legacy-operations/cognify.md): Legacy operation that turns ingested text into a knowledge graph by creating chunks, embeddings, summaries, nodes, and edges in vector and graph stores.
* [Search](https://docs.cognee.ai/core-concepts/main-operations/legacy-operations/search.md): Legacy retrieval operation that blends vector similarity, graph structure, and LLM reasoning to answer questions across one or more Cognee datasets.
* [Memify](https://docs.cognee.ai/core-concepts/main-operations/legacy-operations/memify.md): Legacy enrichment operation that runs configurable extraction and coding pipelines on an existing Cognee knowledge graph to add derived knowledge.
* [Delete](https://docs.cognee.ai/core-concepts/main-operations/legacy-operations/delete.md): Legacy operation to remove data items, datasets, or all permitted data from Cognee, cleaning up graph nodes, vector embeddings, and metadata records.
* [Datasets](https://docs.cognee.ai/core-concepts/further-concepts/datasets.md): Datasets are named containers that group documents and scope ingestion, enrichment, recall, and permissions in Cognee remember and improve flows.
* [NodeSets](https://docs.cognee.ai/core-concepts/further-concepts/node-sets.md): NodeSets are simple tags that group parts of Cognee memory at the dataset level, materialized as graph nodes for scoped retrieval and improvement.
* [Ontologies](https://docs.cognee.ai/core-concepts/further-concepts/ontologies.md): Use an optional RDF/OWL ontology to validate extracted entities, link them to canonical classes, and enrich the Cognee graph with inherited edges.
* [Global Context Index](https://docs.cognee.ai/core-concepts/further-concepts/global-context-index.md): An optional summary layer that builds dataset-level orientation through semantic buckets and a root summary to enrich graph completion retrieval.
* [Agent Memory Decorator](https://docs.cognee.ai/core-concepts/further-concepts/agent-memory-decorator.md): cognee.agent\_memory is an async decorator that retrieves Cognee context before an agent runs and saves its inputs, outputs, and traces afterward.
* [Sessions and Caching](https://docs.cognee.ai/core-concepts/sessions-and-caching.md): How Cognee maintains short-term conversational memory through sessions identified by user and session ID, and the pluggable cache adapter backends.
* [Loaders](https://docs.cognee.ai/core-concepts/further-concepts/loaders.md): Loaders read files from disk or cloud storage and convert them to text, with built-in support for text, PDF, image, audio, CSV, and Office formats.
* [Chunkers](https://docs.cognee.ai/core-concepts/further-concepts/chunkers.md): Chunkers split documents into token-sized pieces for embedding and graph extraction, with built-in text, CSV, LangChain, and code-aware strategies.
* [Multi-User Mode Overview](https://docs.cognee.ai/core-concepts/multi-user-mode/multi-user-mode-overview.md): How Cognee enforces strict data isolation across users and datasets through ENABLE\_BACKEND\_ACCESS\_CONTROL, partitioned graph and vector stores.
* [Dataset Database Handlers: What are they?](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/dataset-database-handlers-what-are-they.md): Dataset Database Handlers are pluggable classes that map each Cognee dataset to its own graph and vector storage backend for full data isolation.
* [Dataset Database Handlers: How to use them?](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/dataset-database-handlers-how-to-use-them.md): How to register, configure, and use Dataset Database Handlers to back Cognee datasets with per-dataset graph and vector storage in multi-user mode.
* [Kuzu Dataset Database Handler](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/existing-dataset-database-handlers/kuzu.md): Use the built-in Kuzu handler to back Cognee datasets with file-based Kuzu graph storage, enabling multi-user mode with zero extra dependencies.
* [LanceDB Dataset Database Handler](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/existing-dataset-database-handlers/lancedb.md): Use the built-in LanceDB handler to back Cognee datasets with file-based LanceDB vector storage, enabling multi-user mode without extra dependencies.
* [PGVector Dataset Database Handler](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/existing-dataset-database-handlers/pgvector.md): Use the PGVector handler to back Cognee datasets with Postgres and the pgvector extension, enabling multi-user mode with shared relational storage.
* [FalkorDB Dataset Database Handler](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/existing-dataset-database-handlers/falkor.md): Use the FalkorDB community handler to back Cognee datasets with FalkorDB graph storage, enabling multi-user mode and per-dataset isolation on Falkor.
* [Neo4j Aura Dataset Database Handler](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/existing-dataset-database-handlers/neo4j-aura-dev.md): Use the Neo4j Aura handler to provision per-dataset Neo4j Aura graph instances for Cognee, enabling cloud-hosted multi-user mode and data isolation.
* [Qdrant Dataset Database Handler](https://docs.cognee.ai/core-concepts/multi-user-mode/dataset-database-handlers/existing-dataset-database-handlers/qdrant.md): Use the Qdrant community handler to back Cognee datasets with Qdrant vector storage, enabling multi-user mode and per-dataset isolation on Qdrant.
* [Permissions System Overview](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/overview.md): Introduction to the Cognee permission system, its core components, and how ENABLE\_BACKEND\_ACCESS\_CONTROL enables multi-tenant data isolation.
* [Datasets](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/datasets.md): Datasets are the dataset-scoped unit of permissioning in Cognee, where every ACL entry, ownership, and access check is anchored at the dataset level.
* [Principals](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/principals.md): Principals are the unified abstraction for users, tenants, and roles in Cognee's permission system, providing one polymorphic mechanism for ACL entries.
* [Users](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/users.md): Users authenticate with email and password, own datasets, inherit permissions from tenants and roles, and are the primary principal type in Cognee.
* [Tenants](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/tenants.md): Tenants represent organizations or groups that contain users and can be granted dataset permissions, applying organization-wide access in Cognee.
* [Roles](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/roles.md): Roles group users within a tenant and can be granted dataset permissions that automatically apply to all members for team-scoped access control.
* [ACL](https://docs.cognee.ai/core-concepts/multi-user-mode/permissions-system/acl.md): The ACL system stores Cognee permissions in the acls table and aggregates principal, tenant, and role entries at runtime to compute effective access.

## Setup & Configuration

* [Setup Configuration](https://docs.cognee.ai/setup-configuration/overview.md): Configure Cognee to use your preferred LLM, embedding engine, relational database, vector store, and graph store via environment variables in a local .env file
* [LLM Providers](https://docs.cognee.ai/setup-configuration/llm-providers.md): Configure LLM providers for text generation, reasoning, and structured output in Cognee using OpenAI, Anthropic, Gemini, or local Ollama and LM Studio models
* [Structured Output Backends](https://docs.cognee.ai/setup-configuration/structured-output-backends.md): Configure structured output backends in Cognee to reliably extract Pydantic models from LLM responses for knowledge graph construction and downstream tasks
* [Embedding Providers](https://docs.cognee.ai/setup-configuration/embedding-providers.md): Configure embedding providers in Cognee to convert text into vector representations for semantic search across OpenAI, local models, and other engines
* [Relational Databases](https://docs.cognee.ai/setup-configuration/relational-databases.md): Configure relational databases like SQLite or PostgreSQL in Cognee to store document metadata, chunk records, pipeline state, and data provenance information
* [Vector Stores](https://docs.cognee.ai/setup-configuration/vector-stores.md): Configure vector databases in Cognee for embedding storage and semantic similarity search across LanceDB, ChromaDB, Weaviate, Milvus, and other backends
* [Graph Stores](https://docs.cognee.ai/setup-configuration/graph-stores.md): Configure graph databases for Cognee knowledge graph storage, entity-relationship reasoning, and traversal across Neo4j, Kuzu, NetworkX, and other backends
* [Permissions Setup](https://docs.cognee.ai/setup-configuration/permissions.md): Configure Cognee permission system, authentication, and access control to enable data isolation and fine-grained authorization across users and datasets
* [Security & Privacy](https://docs.cognee.ai/setup-configuration/security.md): Configure security controls in self-hosted Cognee deployments to protect sensitive data, API keys, network access, and auth for production environments
* [Logging](https://docs.cognee.ai/setup-configuration/logging.md): Understand how Cognee logs work, where files are stored, how to control console and file log levels, and how to debug pipelines with structured log output
* [Adapters Overview](https://docs.cognee.ai/setup-configuration/community-maintained/overview.md): Browse community-maintained adapters that extend Cognee with extra vector stores, graph databases, and providers built and supported by the Cognee community
* [Qdrant](https://docs.cognee.ai/setup-configuration/community-maintained/qdrant.md): Use Qdrant as a Cognee vector store through a community adapter, with cloud-hosted or self-hosted deployments and high-performance embedding similarity search
* [Redis](https://docs.cognee.ai/setup-configuration/community-maintained/redis.md): Use Redis with the Redis Search module as a Cognee vector store through a community adapter for self-hosted setups or managed Redis Cloud deployments
* [FalkorDB](https://docs.cognee.ai/setup-configuration/community-maintained/falkordb.md): Use FalkorDB as a hybrid graph and vector store in Cognee through a community adapter, with cloud-hosted or self-hosted deployments for GraphRAG workloads
* [Memgraph](https://docs.cognee.ai/setup-configuration/community-maintained/memgraph.md): Use Memgraph as an in-memory Bolt-protocol graph store for Cognee through a community adapter for fast knowledge graph traversal and relationship reasoning
* [Pinecone](https://docs.cognee.ai/setup-configuration/community-maintained/pinecone.md): Use Pinecone, a fully managed cloud vector database, as a Cognee vector store through a community adapter for high-performance large-scale similarity search
* [Turbopuffer](https://docs.cognee.ai/setup-configuration/community-maintained/turbopuffer.md): Use Turbopuffer as a Cognee vector store through a community adapter, a high-performance serverless vector database built for fast similarity search at scale

## Guides

* [Permission Snippets](https://docs.cognee.ai/guides/permission-snippets.md): Runnable Python snippets for Cognee's multi-user permissions: create users, tenants, roles, and datasets, then grant and revoke access programmatically.
* [Search Basics](https://docs.cognee.ai/guides/search-basics.md): Query Cognee memory with cognee.recall() and the lower-level cognee.search() API, covering search types, dataset scoping, and permission requirements.
* [Local Setup (No API Key)](https://docs.cognee.ai/guides/local-setup.md): Run Cognee fully offline by pairing local LLMs through Ollama with Fastembed embeddings, avoiding cloud API keys and silent fallbacks to OpenAI.
* [Sessions](https://docs.cognee.ai/guides/sessions.md): Enable conversational memory by reusing a session\_id across recall() calls so Cognee remembers prior Q\&A and answers follow-up questions in context.
* [Agent Memory Quickstart](https://docs.cognee.ai/guides/agent-memory-quickstart.md): Compare session memory and persistent graph memory side-by-side with two minimal agents built on cognee.agent\_memory, remember(), and forget().
* [Deploy REST API Server](https://docs.cognee.ai/guides/deploy-rest-api-server.md): Run Cognee as a REST API server with Docker Compose or a local Python virtualenv to expose ingestion, recall, and graph operations over HTTP endpoints.
* [Time Awareness](https://docs.cognee.ai/guides/time-awareness.md): Build temporal memory with remember() and answer time-bounded questions (before, after, between) using SearchType.TEMPORAL and date-aware recall().
* [Ontology Quickstart](https://docs.cognee.ai/guides/ontology-support.md): Ground Cognee entity and relationship extraction in an OWL ontology (RDF/XML) by attaching the ontology file to a single remember() call for your domain.
* [S3 Storage](https://docs.cognee.ai/guides/s3-storage.md): Ingest data from s3:// URIs with cognee.add() and persist Cognee's data and system roots to S3 or S3-compatible backends like MinIO for shared storage.
* [Graph Visualization](https://docs.cognee.ai/guides/graph-visualization.md): Render your Cognee knowledge graph to a self-contained interactive HTML file with color-coded nodes, labeled edges, drag, zoom, and hover tooltips.
* [Low-Level LLM](https://docs.cognee.ai/guides/low-level-llm.md): Call LLMGateway.acreate\_structured\_output to get Pydantic-validated structured responses from any configured LLM backend without changing your code.
* [Distributed Execution](https://docs.cognee.ai/guides/distributed-execution.md): Scale Cognee pipelines across Modal containers with a one-line toggle to parallelize ingestion and cognify workloads for large batches and slow tasks.
* [Feedback System](https://docs.cognee.ai/guides/feedback-system.md): Capture user ratings on recall answers by attaching feedback to session Q\&A entries with cognee.session.add\_feedback and delete\_feedback APIs.
* [Multilingual Ingestion](https://docs.cognee.ai/guides/multilingual-ingestion.md): Auto-detect languages with langdetect and translate non-English chunks via LLM, Google, or DeepL before Cognee builds the knowledge graph from them.
* [Self-Improvement Quickstart](https://docs.cognee.ai/guides/self-improvement-quickstart.md): Run cognee.improve() to enrich an existing dataset with Memify-style passes that bridge session-only Q\&A content into your permanent knowledge graph.
* [Triplet Embeddings](https://docs.cognee.ai/guides/memify-triplet-embeddings.md): Build or rebuild embeddings over source-relationship-target triplets with the Memify pipeline so SearchType.TRIPLET\_COMPLETION recall returns results.
* [Session Persistence](https://docs.cognee.ai/guides/memify-session-persistence.md): Promote cached conversation Q\&A into the searchable knowledge graph using the lower-level Memify session-persistence pipeline or cognee.improve().
* [Entity Consolidation](https://docs.cognee.ai/guides/memify-entity-consolidation.md): Run the Memify entity-consolidation pipeline to rewrite fragmented descriptions using each node's full graph neighborhood for coherent, complete summaries.
* [Custom Data Models](https://docs.cognee.ai/guides/custom-data-models.md): Define Pydantic DataPoint subclasses and insert typed entities into the knowledge graph with add\_data\_points, skipping cognify and controlling indexing.
* [Custom Graph Model](https://docs.cognee.ai/guides/custom-graph-model.md): Constrain Cognee entity and relationship extraction by passing custom Pydantic graph schemas into remember() for consistent labels, properties, and edges.
* [Custom Tasks and Pipelines](https://docs.cognee.ai/guides/custom-tasks-pipelines.md): Compose Task objects into custom Cognee pipelines that ingest text, extract entities with an LLM, store graph nodes, and visualize the result.
* [Custom Prompts](https://docs.cognee.ai/guides/custom-prompts.md): Shape entity types, relationship labels, and extraction rules by passing a custom\_prompt into cognee.remember() during knowledge graph construction.

## How-To Guides

* [Deployment Overview](https://docs.cognee.ai/how-to-guides/cognee-sdk/deployment.md): Plan a Cognee deployment with flexible relational, vector, and graph storage backends that scale from local development to production.
* [Deployment Options](https://docs.cognee.ai/how-to-guides/cognee-sdk/deployment/deployment-options.md): Compare Cognee deployment patterns by writer ownership, storage backend, and read scaling to pick the right architecture for your workload.
* [Docker Deployment](https://docs.cognee.ai/how-to-guides/cognee-sdk/deployment/docker.md): Run Cognee and supporting services locally or on a server with Docker Compose profiles to start only the containers your workload needs.
* [EC2 Deployment](https://docs.cognee.ai/how-to-guides/cognee-sdk/deployment/ec2.md): Deploy Cognee on Amazon EC2 for full server access, custom configurations, and traditional cloud infrastructure under your direct control.
* [Kubernetes (Helm)](https://docs.cognee.ai/how-to-guides/cognee-sdk/deployment/helm.md): Install Cognee on Kubernetes with the official Helm chart for enterprise-grade, highly available, production-ready container orchestration.
* [Modal Deployment](https://docs.cognee.ai/how-to-guides/cognee-sdk/deployment/modal.md): Run Cognee on Modal for serverless, auto-scaling knowledge graph processing with minimal infrastructure management and pay-per-use billing.

## Examples

* [Overview](https://docs.cognee.ai/examples/overview.md): Pick a Cognee example by use case - vertical AI agents, enterprise data unification, or edge AI - and see the patterns each scenario relies on.
* [Vertical AI Agents](https://docs.cognee.ai/examples/vertical-ai-agents.md): Use Cognee as the memory layer for vertical AI agents, with remember and recall tools that give LangGraph, CrewAI, or custom agents persistent context.
* [Data Silos](https://docs.cognee.ai/examples/data-silos.md): Unify CRM, ERP, support tickets, and documents into a single Cognee memory layer with remember, improve, and recall across 30+ data sources.
* [Edge AI](https://docs.cognee.ai/examples/edge-ai.md): Run the Cognee memory pipeline on-device with cognee-RS, the Rust SDK for offline retrieval on phones, wearables, robots, and smart-home hubs.
* [Cognee Walkthrough](https://docs.cognee.ai/examples/getting-started-with-cognee.md): End-to-end tutorial covering nodesets, ontologies, memify, graph visualization, and a feedback loop using a coding assistant example.
* [Migrate from Mem0](https://docs.cognee.ai/examples/migrate-from-mem0.md): Export Mem0 memories to JSON and re-ingest them into Cognee with remember() so agents keep their long-term context after switching memory backends.
* [Migrate a Relational Database into a Knowledge Graph](https://docs.cognee.ai/examples/relational-db-migration.md): Convert a SQLite or Postgres schema and rows into a Cognee knowledge graph, then ask natural-language questions that span tables and relationships.

## Cognee CLI

* [Cognee CLI](https://docs.cognee.ai/cognee-cli/overview.md): Use the cognee-cli terminal tool to ingest data, run the cognify pipeline, and query AI memory without writing Python, with friendly defaults for beginners.

## Contributing

* [Contributing](https://docs.cognee.ai/contributing/contributing-overview.md): How to contribute to Cognee with bug reports, feature requests, pull requests, tests, and documentation, plus coding standards and review guidelines.

## Additional Docs

* [Changelog](https://docs.cognee.ai/changelog.md): Cognee release notes and version history covering new features, vector and graph store updates, bug fixes, and breaking changes across releases.