The neighbourhood
Friends in the Space
We're glad this space is crowded — it means the problem matters. Giving AI agents durable, useful memory is one of the hard, important problems of this generation, and a lot of talented teams are building toward it from different angles. The tools below are colleagues we genuinely admire — each has shaped how the whole field thinks about memory. We're listing them here not to compare, but to celebrate.
mem0
AI memory that persists across sessions and agents.
- Best-in-class dev ergonomics (pip install, 60k+ stars)
- Production / enterprise maturity (SOC2, HIPAA)
- Serious benchmarking discipline
Zep
Agent memory at enterprise scale.
- Pioneered temporal knowledge graphs
- Sub-200ms p95 at 100M entities
- Provenance-first
Graphiti
Build Temporal Context Graphs for AI Agents.
- Reference implementation for time-aware graphs (Apache-2.0, 29k stars)
- Backend-flexible (Neo4j / FalkorDB / Neptune)
- First-class MCP server
LangMem
The SDK that helps your agents learn.
- Principled cognitive-science taxonomy (semantic / episodic / procedural)
- Novel procedural memory & prompt self-optimization
- Deep LangGraph integration
Letta
Machines that learn.
- Originated the field (MemGPT paper)
- Deepest research bench on continual learning
- Ships as real product
Cognee
Open-Source Memory Platform for Agents.
- Cognitive-science-grounded graph memory
- Pragmatic single-Postgres stack
- Broadest integrations
LlamaIndex
Memory built into the dominant RAG framework.
- Structured memory in the most-used framework
- Composable memory blocks
- Memory + document understanding
Supermemory
The memory layer for AI agents.
- Rare dual dev + consumer focus
- Rich extractors & connectors
- Clean modern API
Kernel Memory (Microsoft)
Memory as a Service.
- Opinionated .NET memory architecture
- Runs four ways
- Genuine upstream influence
All names, logos, and marks are the property of their respective owners.
Listed here with respect, not affiliation.