claude-mem.ai

What It Is

claude-mem is an open-source memory plugin for Claude Code and other AI Agents. CMEM Cloud syncs local memory across different machines, editors, and Agents through a private MCP link. It is positioned not as a chatbot, but as an “Agent memory layer”: while the AI is working, a separate observation-generation mechanism records what happened and builds a searchable, persistent observations database.

Core Capabilities

The core product includes a local SQLite observations DB, semantic vector retrieval, offline-first cloud sync, private MCP links, and Dreaming/live observation generation. According to the official description, observation generation is out-of-band, so it does not block the main Agent, and it uses progressive disclosure to reduce the cost of long sessions. Integrations cover MCP clients such as Claude Code, Cursor, Windsurf, Codex CLI, Gemini, and OpenCode, with example APIs such as GET /memory/recent and POST /memory/observe.

Pricing and Privacy

The open-source engine is currently free to install. CMEM Cloud is in early access, priced at $0 with no credit card required, but users need to join the waitlist to receive an invite. The Team version has not launched yet and is planned to include shared memory, project isolation, role-based permissions, auditing, and priority support; pricing has not been disclosed. On privacy, the official messaging emphasizes local-first: data is stored by default in the user’s local SQLite database, privacy stripping is applied when memories are generated, cloud sync is strictly opt-in, and private MCP links can be revoked and regenerated.

Pros and Cons

Its main advantage is that it packages the database, vector index, sync service, and MCP Server needed for Agent memory, allowing individual developers to try it with a single command. It is also open source and local-first, which reduces lock-in risk. The limitations are also clear: Cloud is still in early access, team governance features have not been formally delivered, and there is no clear information yet on long-term pricing, SLA, compliance certifications, Chinese-language support, or retrieval-quality benchmarks.

Who It’s For and Access from China

It is suitable for developers who frequently use AI coding assistants and want to preserve project knowledge across sessions. It may also appeal to technical teams exploring shared Agent memory. Access from China is not discussed in the main documentation, so real-world usability will require testing network access and MCP endpoint connectivity. No credit card is currently required for payment, and the eventual paid billing methods are unknown. As alternatives, users could consider building their own SQLite/vector database + MCP Server setup, or using the built-in memory and knowledge-base features of existing IDEs or Agents.