Cognee
Verified Oct 23, 2026 ActiveTurns documents and past sessions into a graph-plus-vector memory that AI agents can query across runs.
Listed on EffectHub ·
Pricing
- Pricing model
- Freemium
- Paid options from
- $1/credit
- Billing
- Metered · per credit
Usage-metered rather than a monthly plan: the Standard tier is $1.00 per 1 million tokens processed plus $5 a month per additional workspace. Free covers 1 million tokens, one workspace, unlimited users and API calls without a card; Enterprise is a quoted bring-your-own-cloud engagement, and the open-source SDK is free to self-host.
Use tool ↗About Cognee
Cognee is a memory layer for AI agents. You feed it documents, code or previous conversations, and it builds a knowledge graph alongside vector embeddings so an agent can pull the right context back on a later run instead of re-reading everything. The Python SDK is open source and can run entirely on your own machines, with an MCP server and integrations for Claude Code and Codex. Cognee Cloud is the managed version, adding Slack, Notion, Linear and Google Drive connectors and code indexing for repositories, billed per token processed. Enterprise customers get it deployed in their own cloud as a fixed-scope engagement. It is built by Topoteretes UG in Berlin.
- Platforms
- Web · API
- API
- Yes
- Last verified
- Oct 23, 2026
Verified sources
- Pricing page: Free $0 with 1M tokens and 1 workspace; Standard $1.00 per 1M tokens plus $5 per additional workspace a month, with Slack, Notion, Linear and Google Drive connectors; Enterprise as a BYOC engagement (15 October 2026)cognee.ai ↗
- Pricing page footer: © 2026 Topoteretes UG (haftungsbeschränkt), Schonhauser Allee 163, Berlin, registered at Amtsgericht Charlottenburg; open source, 32.4k GitHub stars (15 October 2026)cognee.ai ↗
Is Cognee legit?
We don’t label tools safe or unsafe. Below are the records we checked ourselves, each with the date we read it.
- Domain registered
- Dec 21, 2023 (2 years ago)Registry record (RDAP), registrar NameCheap, Inc., looked up Sep 17, 2026
- TLS certificate
- Issued by Let's Encrypt, valid until Jan 14, 2027Seen in our own connection to the site on Oct 23, 2026
- Last checked
- Oct 23, 2026The site answered
- On the market since
- 2024The vendor's own material, read by us
- Operated by
- Topoteretes UG (haftungsbeschränkt), GermanyOur research, card verified Oct 15, 2026
Pros & cons of Cognee
Cognee's bet is that plain vector search loses the relationships between facts, so it builds a graph on top and lets agents traverse it. Being open source means you can run the whole thing locally and pay only if you want the hosted version, which is the right shape for infrastructure. The catch is that graph construction costs tokens, so a large corpus is not free even at $1 per million, and you are wiring a library rather than clicking through a product.
Pros
- Open-source Python SDK you can run entirely on your own machines
- Builds a knowledge graph alongside embeddings instead of only vectors
- MCP server plus integrations for Claude Code and Codex
- Free tier and pay-per-token pricing rather than a seat licence
Cons
- Ingesting a large corpus burns tokens, and per-million billing adds up on big re-indexes
- It is a library first, so expect to write code rather than configure a dashboard
- Workspaces are billed separately at 5 dollars each beyond the first
Best for
- Developers building agents that need recall across sessions
- Teams whose data has structure that flat embeddings lose
- Anyone who wants to self-host the memory layer rather than rent it
Not for
- Non-developers looking for a finished application
- Projects happy with a plain vector store and a single retrieval hop
Watch out: The price is per million tokens processed, not a monthly subscription. Re-indexing a big repository is a billable event.
Alternatives to Cognee
Cognee head-to-head
Side-by-side fact sheets against the competitors our research or their makers name — pricing, platforms, API and both verdicts on one page.