memorygraph.dev

What It Is

MemoryGraph is a persistent memory system built specifically for AI assistants. Built on MCP (Model Context Protocol), it stores memories using a knowledge graph rather than a simple vector database, allowing AI to remember conversations, code patterns, and solutions across sessions while understanding causal relationships between memories.

Core Dimensions

AI capabilities and models: MemoryGraph is not a large language model itself, but memory middleware. Its main strength is using a graph structure—with typed edges such as SOLVES and DEPENDS_ON—to preserve context and causality. Compared with flat vector retrieval, it can answer “what solved X” more precisely, rather than merely “what is similar to X.”
API and integrations: It is deeply integrated with the MCP ecosystem, supporting Claude Code/Desktop, Cursor, VS Code Copilot, and more. It also offers a Cloud REST API, with Python/Node.js SDKs coming soon, and is compatible with mainstream frameworks such as LangChain and LlamaIndex.
Data privacy: It follows local-first and privacy-by-default principles. SQLite is used by default, data is encrypted both in transit and at rest, the company promises not to use data for model training, and full self-hosting is supported.
Output quality and limitations: Citing DeepMind benchmark results, its graph-based memory significantly outperforms pure vector-based solutions such as Mem0 in reusing past experience. However, it is not an autonomous agent: users need to manually configure “memory instructions” in places such as CLAUDE.md for storage and recall to be triggered effectively. In addition, the cloud backend is still in a waitlist stage.

Pricing

The open-source MCP server is free forever. The cloud service has three tiers: Pro at $5/month for 100,000 memories, Ultra at $50/month for 500,000 memories, and Team at $100/month for unlimited memories and 10 users. Self-hosting is completely free, making it very developer-friendly.

Pros and Cons

Pros: high-fidelity graph memory, strong local-first privacy, open source and self-hostable, and seamless backend data migration. Cons: manual instruction setup creates a modest barrier to entry, the cloud service is not fully launched yet, and the best automated experience depends on the Claude Code plugin.

Who It’s For

Developers who rely heavily on AI coding tools, teams that need AI to maintain context across sessions, and users with high data privacy requirements who also have some technical background.

Access from China

The official .dev domain and API domain usually require a proxy to access. Payment methods are unclear, but will most likely require an international credit card. An alternative is Mem0, which is mentioned in the article.