Three-stage ECL pipeline: Extract ingests 30+ multimodal formats (PDF, Slack, Notion, images, audio) via dedicated connectors with auto-OCR and transcription; Cognify runs LLM-powered entity/relationship extraction into subject-predicate-object triples with RDF/OWL ontology alignment and URI-based coreference resolution (unifying name variants across documents into canonical identifiers); Load dual-writes to a pluggable graph store (Postgres native, Neo4j, Kuzu) and vector store (pgvector, Qdrant, LanceDB). Retrieval combines graph BFS traversal, vector similarity, temporal filtering, and auto-routing query optimization; hot context is cached in session memory with transparent fallback to the persistent knowledge graph across sessions.
- Storage
- Dual-write: pluggable graph store (Postgres native, Neo4j, Kuzu, Neptune) and vector store (pgvector, LanceDB, Qdrant, ChromaDB, Weaviate); relational metadata via SQLAlchemy. Local self-host or managed cloud.
- Retrieval
- Graph BFS traversal, vector similarity, temporal filtering, and auto-routing query optimization over a persistent ontology-grounded knowledge graph; session memory cache falls through to the permanent graph for cross-session recall; agentic skill-based retrieval also supported.
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Free (1M tokens/mo included, 1 workspace); Standard $2.50/1M tokens + $5/additional workspace (unlimited workspaces, Slack/Notion/Drive integrations); Enterprise custom (BYOC, SLA, dedicated support). The former 'Hobby'/'Growth' tier names were renamed to 'Free'/'Standard' on the pricing page; terms unchanged. · Free + paid
- GitHub stars
- 29,917
- Last release
- 2026-08-08
- Last commit
- 2026-08-10
- First catalogued
- 2026-06-28
Strengths
- Cognify stage extracts structured entity/relationship triples with RDF/OWL ontology alignment and URI-based coreference resolution, eliminating duplicate-entity drift across ingestion runs
- Multi-hop graph reasoning via BFS traversal over the ontology-grounded knowledge graph — outperforms pure vector RAG on multi-hop QA benchmarks
- Pluggable backends: swap graph store (Postgres, Neo4j, Kuzu) or vector store (pgvector, Qdrant, LanceDB, Weaviate) without changing application code
- First-party adapters for LangGraph and CrewAI alongside MCP server (stdio + HTTP), REST API, CLI, and Python SDK
- Session memory with background graph synchronization: hot context cached locally and persistent knowledge graph updated asynchronously across sessions
- v1.3.0 (Jul 2026) adds an optional Topic Index that clusters ingested documents and generates per-cluster overviews, plus metadata filtering (tags/author/date ranges) and configurable embedding model/chunk size at ingest time
Watch out
- No SOC2/HIPAA as of mid-2026 — disqualifying for regulated data
- RDF/OWL ontology alignment in the Cognify stage requires semantic-web expertise; teams without this background face a steep learning curve when customizing or debugging graph extraction pipelines
- Graph BFS traversal is slower than pure vector RAG; TB-scale deployments require manual Neo4j sharding and vector store cluster tuning — self-host operational burden grows substantially at production scale
Best for
- Cross-document multi-hop reasoning where entity identity must be preserved across ingestion runs (legal, finance, research evidence graphs)
- LangGraph or CrewAI agents needing first-party persistent memory without building a custom graph layer
- Enterprise knowledge-graph workflows over large mixed-format corpora where ontology-backed deduplication matters
How it integrates
Benchmark results
| Benchmark | Value | Backbone | Trust | Source |
|---|---|---|---|---|
| beam-10m | 0.67 accuracy | — | Self-reported | cognee maintainers (README Benchmarks section) ↗ |
Sources
- Pricing page re-checked — unchanged: Free $0/mo (1M tokens, 1 workspace); Standard $2.50/1M tokens plus $5 per additional workspace; Enterprise contact-us (dedicated Slack, dedicated support engineer, BYOC, SLA) (vendor)
- Monthly card re-verification — GitHub API repo metadata: 29,656 stars, license Apache-2.0 confirmed, archived=false, last commit on the default branch 2026-07-31 (third-party)
- https://www.cognee.ai/ (vendor)
- Cognee GitHub — 25.7k stars, Apache-2.0 license, v1.2.2 latest release (vendor)
- Cognee pricing — Hobby free/Growth $5/workspace/mo/Enterprise custom; $2.50/1M tokens across all tiers (vendor)
- Cognee releases — v1.2.2 (2026-06-26) Truth Subspace & Retrieval Improvements; v1.2.0 session distillation + proposals API (vendor)
- Cognee — v1.0, Rust edge engine, enterprise use (Bayer, U. Wyoming) (vendor)
- Official MCP server — stdio (default) and HTTP transport; remember/recall/forget tools (vendor)
- REST API deployment docs — FastAPI server on localhost:8000; Docker Compose option (vendor)
- Python SDK on PyPI — pip install cognee; Python 3.10–3.14 (vendor)
- CLI reference — cognee-cli ships with the cognee pip package; remember/recall/improve/forget/push/-ui (vendor)
- First-party LangGraph adapter — Cognee store/search as @tool-decorated LangGraph-compatible functions with session isolation (vendor)
- First-party CrewAI adapter — Cognee knowledge store as CrewAI-compatible tools with session isolation (vendor)
- Cognee architecture — vector backends (LanceDB, Chroma, pgvector, Qdrant, Weaviate), graph backends (Postgres, Neo4j, Kuzu), ECL pipeline stages (third-party)
- GitHub API star re-check — 26,107 stars (vendor)
- GitHub API re-check — 27,949 stars, pushed 2026-07-16 (third-party)
- Cognee v1.3.0 'Smarter Search & Dataset Indexing' (2026-07-12) — optional Topic Index (document clustering + per-cluster overviews), batch/resumable ingestion, metadata filtering in search, configurable embedding model + chunk size, search ranking improvements; no breaking changes (vendor)
- GitHub API re-check — license still Apache-2.0, not archived, pushed 2026-07-23T17:18Z (third-party)
- Cognee pricing re-check — tier names changed from 'Hobby'/'Growth' to 'Free'/'Standard'; prices and limits unchanged ($0 with 1M tokens incl.; $2.50/1M tokens + $5/workspace; Enterprise custom) (vendor)
- Cognee v1.4.0 'Search & Ingestion Improvements' (2026-07-17) — faster/more reliable ingestion for large batches, search-relevance tuning, an optional dataset-level overview index (extends the v1.3.0 Topic Index concept), dataset-management API additions; no breaking changes per release notes (vendor)
Last verified 2026-08-01 · updated by refresh-framework-cards