Claude-Mem
Persistent AI memory for Cursor (and Claude Code) — session hooks capture tool use, AI compresses observations, and relevant history injects into every new chat.
Overview
Claude-Mem (thedotmack/claude-mem) fixes cold-start sessions. Hooks capture MCP tools, shell commands, and file edits; an extraction layer compresses observations into semantic summaries; context injection loads relevant history into new chats. A memory viewer shows the knowledge base. Cursor-only users can run on Gemini’s free tier or OpenRouter — no Claude Code subscription required. Claude Code users install via plugin marketplace + `claude-mem cursor install`.
Work with LimeDock
Using this skill? LimeDock can wire it into a durable automation you own.
Skills show what's possible. LimeDock builds and runs the owned marketing, sales, and ops automations around them.
We sell owned automations for SaaS teams — live workflows that plug into Slack, CRM, and your internal platform — not just a skill list.
Link
Installation guide
# Cursor-only (Gemini / OpenRouter)
git clone https://github.com/thedotmack/claude-mem.git
cd claude-mem && bun install && bun run build
bun run cursor:setup
# Claude Code users
/plugin marketplace add thedotmack/claude-mem
/plugin install claude-mem
claude-mem cursor installPrereqs: Bun, Cursor IDE, jq + curl. Verify with `bun run worker:status` and `bun run cursor:status`.
How to use it
Keep the worker running. Code normally — hooks capture silently. New chats should receive relevant prior context. Open the worker URL to browse memory. Restart Cursor after install.
Example prompts
- “After install: continue yesterday’s auth refactor — what did we decide about session tokens?”
- “Show me what claude-mem has stored about our billing module.”
- “Summarize architecture decisions from the last five sessions.”
Use cases and examples
- Stop re-explaining project conventions every session
- Accumulate architecture decisions across weeks
- Cursor workflows without losing multi-day context
- Team demos of persistent agent memory vs file-dump approaches
Prerequisites
- Bun runtime + Cursor IDE
- AI provider: Gemini free tier, OpenRouter, or Claude SDK
- Disk + local worker port available
Tips
- Start with Gemini free tier (1500 req/day) for individuals.
- If hooks don’t fire: restart Cursor and check `.cursor/hooks.json`.
- Complement with file-based CLAUDE.md / knowledge graphs — memory tiers stack.
- Useful backdrop: NexusTrade / DEV.to pieces on Cursor vs Claude Code memory architecture.
Sources
- https://docs.claude-mem.ai/cursor
- https://github.com/thedotmack/claude-mem
- https://nexustrade.io/blog/cursor-vs-claude-code-memory-architecture-20260413
- https://dev.to/austin_starks/cursor-beats-claude-code-heres-the-memory-architecture-that-proves-it-e1
- https://www.datacamp.com/tutorial/claude-code-in-cursor
- https://www.spacecake.ai/blog/cursor-claude-integration-guide