Guide · checked 12 November 2026
memgit vs Claude Code memory, Mem0 and claude-mem
All four give an AI coding agent memory between sessions. They differ in where that memory lives, which tools can read it, and what happens when you move to another machine. Every statement about another product below comes from that product's own documentation, pricing page or README, linked at the end, as read on 12 November 2026.
| Claude Code built-in | Mem0 | claude-mem | memgit | |
|---|---|---|---|---|
| Where memory lives | Markdown files on this machine (CLAUDE.md, and auto memory under ~/.claude/projects/<project>/memory/) | In your Mem0 account when used through its plugin or hosted MCP server; self-hosting is also offered | SQLite at ~/.claude-mem/claude-mem.db, with Chroma for semantic search | A local store on your machine; a copy on memgit cloud only if you turn sync on, encrypted before it leaves |
| Other tools | Claude Code; it can import AGENTS.md and other tools' rules files | Claude Code through its plugin; its docs also describe a hosted MCP server for Cursor and Codex | Claude Code, Codex, Cursor, OpenCode and 11 more agents, per its README | Claude Code, Cursor, Codex, Windsurf, Gemini CLI and any MCP host, one shared store |
| Across machines | No. Anthropic's docs: auto memory is machine-local and not shared across machines | Yes, because memories are stored in the hosted account | Opt-in cloud sync | Opt-in end-to-end encrypted sync; the server stores ciphertext only |
| History and undo | Plain files you can edit or delete; the memory page describes no history | A forget command in the plugin; its docs describe no undo | Not described on the pages read | Every change is a checkpoint: memgit diff shows it, memgit rollback restores it |
| Model calls | Part of your Claude Code session | Uses an LLM and an embedding model (defaults named in its README) or Mem0's platform | Compresses captured tool calls with an AI observer model; provider usage may cost extra | None. Recall is BM25 search over typed memories, no API key |
| Price | Included with Claude Code | Free tier with monthly add and retrieval limits; usage beyond it is billed | Engine free and open source; CMEM Pro $30 a month, card required for its trial | Free and MIT locally; sync is $12 a month or $99 a year, 14 days free without a card |
When the built-in memory is enough
If you use Claude Code on one machine and no other agent, the built-in memory already remembers what it learns. memgit sits on top of it rather than replacing it: memgit init imports those files, so nothing you have built up is lost if you add memgit later.
Hosted memory or a local store
Mem0 keeps memories in its hosted platform when you use its plugin, which is what lets them follow you, and it needs a Mem0 account and API key. claude-mem and memgit keep memory on your machine and offer sync as an option. memgit encrypts on your machine before sync, so the memgit cloud server holds ciphertext it cannot read.
History is the difference you notice later
An agent that remembers will eventually remember something wrong: an outdated build step, a rule it overwrote. memgit records every change as a checkpoint, so you can see what changed and put it back:
memgit log -n 5 memgit diff <checkpoint> memgit rollback <checkpoint>
Try memgit
pip install memgit memgit init
Inside Claude Code, the plugin does the same: /plugin marketplace add code4161/claude-plugins, then /plugin install memgit@memgit. Setting it up on more machines is in memory on every machine.
Sources, read 12 November 2026
- Claude Code docs, memory
- Mem0 pricing, Mem0 for Claude Code, Mem0 MCP, Mem0 on GitHub
- claude-mem on GitHub, CMEM pricing
Products change. If something here is out of date, write to hello@memgit.dev and it will be corrected.