Mem0 alternatives: pick by what you need
There is no single best memory layer, so this guide starts from the requirement. Find the row that matches your constraint, follow it to the options, then read the detail pages. Every fact below is from the project’s own docs or repository.
Reviewed October 2026
Start with your constraint
Managed memory: Mem0 Platform, Zep Cloud, Cognee Cloud, Supermemory’s cloud, Hindsight Cloud and Letta Cloud all host it for you.
Mem0’s self-hosted server (Apache 2.0) needs Postgres with pgvector and an LLM provider. Hindsight (MIT) runs on Postgres with an LLM provider. Supermemory has a local single-binary server, licensed up to 10,000 documents. Dakera is one self-hosted binary with the SDKs under MIT and a proprietary server.
Zep’s open-source route is Graphiti, a framework for temporal context graphs that needs an LLM and a graph database (Neo4j, FalkorDB or Neptune). Cognee builds a knowledge graph from documents, code and conversations.
Dakera runs embeddings, reranking and entity extraction on-device, so there is no per-call model cost. Cognee’s local mode ingests and retrieves text without an LLM API key.
Dakera v0.12.0 adds bge-m3 embeddings for 100+ languages, on-device speech to text and visual recall over document pages. Supermemory pulls in documents through connectors to Drive, Gmail and Notion. Mem0 open source lets you choose your embedder, so language coverage follows that choice.
LangMem is an MIT-licensed library that stores through the LangGraph store, which keeps memory inside your LangGraph process.
Letta is an agent harness: the model manages its memory as part of the agent loop, in Letta Code and Letta Cloud.
The options at a glance
| Project | Shape | Licence |
|---|---|---|
| Mem0 | Managed platform, open-source library, self-hosted server | Apache 2.0 |
| Zep | Zep Cloud; Community Edition is retired, Graphiti is the open-source route | Graphiti: Apache 2.0 |
| Letta | Agent harness (Letta Code) and Letta Cloud | Apache 2.0 |
| Cognee | Open-source memory platform and Cognee Cloud | Apache 2.0 |
| LangMem | Library for LangGraph agents | MIT |
| Supermemory | Cloud service and local single-binary server | MIT (repository) |
| Hindsight | Memory system on Postgres, plus a hosted cloud | MIT |
| Dakera | One self-hosted binary; Dakera Cloud coming | SDKs, CLI and MCP server MIT; server proprietary, no usage fees |
Where Dakera fits
Dakera suits teams who want memory on their own infrastructure without assembling services: hybrid retrieval with a cross-encoder reranker, a knowledge graph, six decay curves, sessions and namespaces, MCP tools and a dashboard, all in one process with its models baked in. Python, TypeScript, Go and Rust SDKs cover the common stacks. Details are on the Dakera vs Mem0 and self-hosting pages.
On LoCoMo, Dakera reports 88.2% Recall@20 (does retrieval surface the answer); vendors that publish LoCoMo scores report QA accuracy (does an LLM answer correctly). Different measurements; see the benchmark methodology.
Frequently asked questions
What is the best Mem0 alternative?
It depends on the constraint. For managed hosting, look at Zep Cloud, Supermemory or Hindsight Cloud. For a graph, Graphiti or Cognee. For self-hosting with no model providers to connect, Dakera. The buyer’s guide above maps each need to the options.
Which Mem0 alternatives can I self-host?
Mem0 itself (library and server), Hindsight, Supermemory’s local server, Cognee, Graphiti and LangMem as libraries, and Dakera as a single binary. Their dependencies differ: Postgres, a graph database, an LLM provider or none.
Which alternatives avoid an LLM provider?
Dakera runs embeddings, reranking and entity extraction on-device, and Cognee’s local mode works without an LLM API key for text ingestion and retrieval. Mem0, Graphiti and Hindsight use an LLM provider for extraction.
Is Zep Community Edition still available?
Zep has retired Community Edition and moved it to a legacy folder. Its repository points to Zep Cloud and the open-source Graphiti library. See Dakera vs Zep.
Run Dakera on your own infrastructure.
One binary with its models baked in and no per-call fees. Start it with one Docker command and see it for yourself.
Prefer managed hosting? Join the Dakera Cloud waitlist →