Cognee
Also known as: topoteretes cognee
Open source AI memory platform giving agents persistent long term memory through a hybrid graph, vector, and relational engine with a memory native API.
Cognee is an open source AI memory platform that gives agents persistent long term memory across sessions. Instead of a passive vector store, it combines vector embeddings, graph reasoning, and relational storage into a single engine, and continuously builds a self hosted knowledge graph through an Extract, Cognify, Load pipeline that ingests data from more than thirty sources and structures it into entities, relationships, and ontology grounded meaning. A memify layer then refines the graph through feedback, so rated responses feed back into edge weights and the memory gets sharper with use. The company raised a seven and a half million dollar seed round in early 2026, is independent, and runs live in more than seventy companies including Bayer and the University of Wyoming, with over a million pipeline runs a month and more than twelve thousand GitHub stars.
The developer facing surface is a memory native API built around four operations, remember, recall, forget, and improve, available through Python and TypeScript SDKs, a CLI, and the Model Context Protocol so any MCP compatible agent framework can use Cognee as a shared memory service. It integrates natively with LangGraph, CrewAI, the Claude Agent SDK, the OpenAI Agents SDK, Google ADK, and n8n, and separates short term session memory for fast reasoning from permanent long term knowledge, with tenant and user isolation so multiple agents can share one graph while keeping their own write scope.
Cognee runs embedded with sensible defaults, then swaps in Neo4j, Neptune, Qdrant, or pgvector when a team scales, and a Rust engine targets on device and edge memory where latency and privacy matter. Pricing keeps the open source core free to self host, adds a free cloud tier with one workspace and a million tokens included and no card required, a usage based Cloud Pro at two dollars and fifty cents per million tokens, and an Enterprise tier with a private cloud, bring your own key, dedicated support, and SLAs. Full data ownership and deployment flexibility are the core promise.
Vendor details
Canonical URL
https://www.cognee.ai
Category
Agent infrastructure
Subcategory
Memory
Funding status
Independent. Raised a $7.5 million seed round in early 2026. Open source (Apache 2.0) with more than twelve thousand GitHub stars and 80 plus contributors. Runs live in more than seventy companies including Bayer and the University of Wyoming, processing over a million pipeline runs a month.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Memory native API (remember, recall, forget, improve) through Python and TypeScript SDKs, a CLI, and the Model Context Protocol. Native integrations with LangGraph, CrewAI, the Claude Agent SDK, the OpenAI Agents SDK, Google ADK, and n8n, ingesting from more than thirty data sources. Pluggable storage across Neo4j, Neptune, Qdrant, and pgvector.
In practice
Your agent forgets everything when a session ends. You add Cognee's remember and recall API, and it builds a persistent knowledge graph so the agent carries context, preferences, and decisions across sessions.
You need multi hop reasoning over connected documents, not flat chunks. Cognee's Cognify pipeline extracts entities and relationships into a graph, so retrieval traverses relationships instead of nearest neighbor lookups alone.
Regulated data cannot leave your infrastructure. You self host the open source engine on your own Neo4j and Qdrant, keep full data ownership, and still give every MCP agent one shared memory.
Sources & related URLs
Capability coverage
6.5 / 14 capabilities · 46%
| Integrations & Tool CallingIntegrates LangGraph, CrewAI, Claude Agent SDK, OpenAI Agents SDK, Google ADK, n8n, MCP, 30 plus sources, docs 2026-07-06 | Full |
|---|---|
| Workflow OrchestrationMemory layer, not an orchestrator | Unable to verify |
| Knowledge Grounding & RAGCore capability: hybrid graph and vector GraphRAG with ontology and multi hop retrieval, docs 2026-07-06 | Full |
| Human Oversight & GuardrailsFeedback loops improve memory but no guardrails or human oversight product | Unable to verify |
| Security, Identity & GovernanceTenant and user isolation, data residency, self host and on prem, audit traits and OTEL; enterprise identity less documented, docs 2026-07-06 | Partial |
| Observability & AuditabilityTraceability, OTEL collector, and audit traits, docs 2026-07-06 | Partial |
| Memory & State PersistenceCore product: persistent long term memory across sessions with session and permanent layers, docs 2026-07-06 | Full |
| Deployment & Data ResidencyManaged cloud, self host, on prem, and edge via a Rust engine with data residency, docs 2026-07-06 | Full |
| Prebuilt Agents, Templates & PacksRetrieval modes and plugins exist but no prebuilt agents | Unable to verify |
| Triggers & Channel CoverageNo event triggers or channel coverage | Unable to verify |
| Model Flexibility & RoutingProvider agnostic, free to run with the LLM and embedding providers you choose, no routing, docs 2026-07-06 | Partial |
| APIs, SDKs & MCP ExtensibilityPython and TypeScript SDKs, CLI, memory native API, MCP, pluggable databases, docs 2026-07-06 | Full |
| Testing, Debugging & OptimizationPublishes memory benchmarks but is not a user facing eval or testing product | Unable to verify |
| Browser & Computer UseNo browser or computer use | Unable to verify |
Pricing
Open source self host free; Cloud Free ($0, 1M tokens); Cloud Pro $2.50 per 1M tokens; Enterprise
tokens processed
Included quota
Open source self host is free with the full engine. Cloud Free includes one workspace and 1 million tokens a month with no card. Cloud Pro is usage based at $2.50 per 1 million tokens. Enterprise adds a private cloud, bring your own key, dedicated support, and SLAs.
What is public
The free tier, the $2.50 per 1 million token Cloud Pro rate, and the open source self host path are clearly published.
Billing mechanics
Open source and free cloud tiers carry the engine at no vendor cost. Cloud Pro meters tokens processed at $2.50 per million. Enterprise runs in a private cloud with bring your own key so model and infrastructure costs stay on the customer's accounts.
Cost watchouts
Persistent memory trades an upfront ingestion token cost for cheaper queries, with break even around two dozen repeated queries, so low reuse workloads pay more per answer. Running your own graph and vector databases at scale adds operational cost.
Variable cost rationale
Token based cloud pricing scales with ingestion and query volume, but the free tier and modest $2.50 per million rate keep growth predictable, and self hosting caps the vendor cost entirely.
Overage / add-ons
Cloud usage bills at $2.50 per 1 million tokens processed above the free allowance.
Sales call required
Mixed (some tiers require a call)
Free / trial
Open source free to self host, plus a free cloud tier with 1 million tokens and no card
Lowest paid plan
Cloud Pro at $2.50 per 1 million tokens, or free via self hosted open source
Commercial notes
Independent. $7.5 million seed in early 2026. Apache 2.0 open source with 12,000 plus GitHub stars. Deployments include Bayer and the University of Wyoming.
Key ambiguities
Cloud SaaS is in beta; Enterprise pricing is scoped with the vendor.
Related vendors
- AgentOps — Agent observability and reliability platform with broad model and…
- Agno — High-performance agent runtime and framework (formerly Phidata) with…
- Apify — Cloud platform for web scraping and automation with 45,000+ prebuilt…
- Arcade — Authenticated tool calling platform and MCP runtime that handles…
- Arize AI — AI observability and evaluation platform that traces, evaluates, and…
- Beam — Open source serverless sandboxes and GPU cloud for production…