Blog
claude-mem alternatives for Codex, Cursor and teams
Published
If you want claude-mem for Codex, Cursor or OpenCode, you may not need an alternative: as of 13 November 2026 its own installer wires all three. Look elsewhere for a different reason. For memory you can read and edit as Markdown, Basic Memory. To run everything on your own machine with a knowledge graph, Cognee. For hosted automatic memory with a team plan, Supermemory. To drop the plugin and use what your agent already has, its built-in memory. And if what you need is for one person's lesson to reach every teammate's agent, through a gate you set, that is a team knowledge layer such as Harbor.
We make Harbor, one of the options below. Everything about the other tools is from their own GitHub pages, docs and pricing, read on 13 November 2026. What claude-mem is and how it works is in what is claude-mem.
Does claude-mem work with Codex, Cursor and OpenCode?
Yes. The claude-mem README (version 15.18.3) says it works with Claude Code, Codex, Cursor, OpenCode and 11 more agents, and its installation guide lists the hosts npx claude-mem install detects: Claude Code, Cursor, Windsurf, OpenCode, Codex CLI, Antigravity CLI, Grok Bot and OpenClaw. Some agents, such as Copilot CLI, can search the memory but do not have their sessions recorded.
So the search “claude-mem alternative for Codex” is usually a few months out of date. Run the installer, pick the hosts you use, and every agent writes to and reads from the same ~/.claude-mem on your machine.
Why people look for a claude-mem alternative
- You cannot read what it remembered. Memory is compressed observations in SQLite and a Chroma index. You search it, or browse it in the local viewer, but it is not a set of files you open and edit.
- It records everything, and you review nothing. Hooks capture sessions and a model writes the summary. Wrap text in
<private>tags to keep it out; otherwise, what a session got wrong is stored with what it got right. - The hosted observer is a subscription. The installer offers CMEM Pro with a 30-day free trial; afterwards memory falls back to your own Anthropic plan unless you subscribe. Anonymous usage analytics are on by default (
npx claude-mem telemetry disable). - You need it for a team. TeamBrain, its team workspace, shares the work lists teammates choose to share, per project, through a hosted cmem.ai account. Its docs describe no review step before a teammate sees what was shared, and it shares work, not the rules a team has agreed on.
1. Basic Memory: memory you can read
Markdown files that you and your agents both write, over an MCP server that its README lists for Claude Code, Codex, Cursor, VS Code and Claude Desktop. A Claude Code plugin adds session briefings on top of the MCP server.
- Instead of claude-mem because: every memory is a plain file you can open in Obsidian or your editor, fix, or delete.
- Watch out for: nothing is captured for you. The agent or you writes each note.
- Price: free locally under AGPL-3.0. The cloud is $15 a month after a 7-day trial; Basic Memory Teams shares one workspace.
2. Cognee: a local knowledge graph
Open-source graph memory under Apache-2.0. Its Codex and Claude Code plugins share one configuration and, with no server set, start a local Cognee API on your machine. In local mode it can route its own model calls through Claude Code.
- Instead of claude-mem because: memory is a graph of entities and relations rather than a list of observations, and Claude Code and Codex share it by default.
- Watch out for: capture is automatic and no review step is documented, the same trade as claude-mem.
- Side by side: Harbor vs Cognee.
3. Supermemory: hosted, with a team plan
One command sets up memory in Claude Code, Cursor, OpenCode and Codex (changelog, 20 June 2026). The engine runs on Supermemory's servers.
- Instead of claude-mem because: nothing runs on your machine, and the Pro plan includes three team seats.
- Watch out for: it is hosted unless you are on a self-hosting plan, and stated facts go in without review; only facts it inferred wait for someone to approve them.
- Price: free with $5 of credits a month; Pro is $19 a month (pricing).
- Side by side: Harbor vs Supermemory.
4. Your agent's built-in memory
Each agent now keeps some memory without a plugin. It is per person and per machine, like claude-mem, but there is nothing to install.
- Claude Code: auto memory, on by default, writes a
MEMORY.mdindex per repository on your machine. See Claude Code auto memory. - Codex: memories are off by default; turn them on with
memories = trueunder[features]. They stay in~/.codex/memories/. See Codex memories. - Cursor: removed its Memories feature in version 2.1. What persists is your rules and AGENTS.md. See does Cursor have memory.
If claude-mem was mostly saving you from repeating preferences, built-in memory plus a short CLAUDE.md or AGENTS.md may be enough.
5. Harbor: when the memory has to reach the team
Harbor is not a drop-in replacement. claude-mem remembers what your sessions did; Harbor serves what your team decided. It learns rules from pull request reviews, Slack and docs, plus what agent sessions record, and gives each Claude Code, Cursor, Codex or Copilot session the ones that apply to its task. You set the gate: Manual holds everything for a person, Hybrid (the default) puts routine rules live and holds the rest, and a clash Harbor cannot settle always waits for a person. The same rules are checked on every pull request.
- Instead of claude-mem because: one person's correction reaches every teammate's agent, in any of the four agents, only after it passes the gate you set. Each rule shows how often it was served and how often an answer cited it (which rules got used).
- Watch out for: it is hosted only, it is new (public sign-up opened on 31 October 2026), and it does not record your sessions for your own recall the way claude-mem does.
- Price: Free costs nothing for up to 2 people. Pro is $24 per person a month, billed yearly.
The two can run side by side: claude-mem for your own continuity, Harbor for what every agent on the team must know. How teams share Claude Code memory today, with and without a tool, is in Claude Code team memory.
Side by side
| Claude Code | Codex | Cursor | Who writes it | Review before sharing | Runs locally | |
|---|---|---|---|---|---|---|
| claude-mem | Hooks | Installer | Installer | Hooks and a model | Not documented | Yes, Apache-2.0 |
| Basic Memory | MCP and plugin | MCP | MCP | You or the agent | No approval step | Yes, AGPL-3.0 |
| Cognee | Plugin | Plugin | MCP | Automatic | Not documented | Yes, Apache-2.0 |
| Supermemory | Plugin | Plugin | Plugin | Automatic | Inferred facts only | Self-hosting plans only |
| Built-in memory | Auto memory | Memories (opt in) | Rules only | The agent | No | Yes |
| Harbor | Hooks, MCP | Hooks, MCP | Hooks, MCP | Reviews, Slack, docs, sessions | A gate you set | No, hosted |
How to choose
- You only wanted claude-mem in Codex, Cursor or OpenCode: keep it and run
npx claude-mem install. - You want to read and fix what was remembered: Basic Memory.
- You want a graph, on your own machine, shared by Claude Code and Codex: Cognee.
- You want nothing running locally and a few team seats: Supermemory.
- You were saving yourself from repeating preferences: your agent's built-in memory and a short rules file.
- You want what one teammate learns to reach everyone's agent, through a gate, and checked on pull requests: Harbor.
Leaving Mem0 instead? See Mem0 alternatives for coding agents. Every tool compared with Harbor is on the comparison pages. If a line here is out of date, tell us and we will fix it.
Questions
Does claude-mem work with Codex and Cursor?
Yes. As of 13 November 2026 its README says it works with Claude Code, Codex, Cursor, OpenCode and 11 more agents, and npx claude-mem install detects and wires Claude Code, Cursor, Windsurf, OpenCode, Codex CLI and others. Some agents, such as Copilot CLI, can search the memory but do not have their sessions recorded.
What is the best claude-mem alternative?
It depends on why you are leaving. For memory you can read and edit as Markdown, Basic Memory. For a local knowledge graph shared by Claude Code and Codex, Cognee. For hosted memory with team seats, Supermemory. To skip a plugin, your agent's built-in memory. For a team's rules in every teammate's agent through a gate you set, Harbor.
Can a team share claude-mem?
Through TeamBrain, which shares the work lists teammates choose to share, per project, via a hosted cmem.ai account. Its docs describe no review step before teammates see what was shared.