Blog
Supermemory alternatives for coding agents (2026)
Published: November 9, 2026
We make Harbor, one of the options below. Everything here about Supermemory and the other tools is from their own sites, docs and repositories, read on 9 November 2026.
The short answer depends on what you want your agents to remember. To keep automatic session memory but run it yourself, Cognee or claude-mem. For facts that change over time, Zep. For an agent that owns its memory, Letta. For memory as files you own, ByteRover. For a few stable rules, a committed CLAUDE.md or AGENTS.md. And for one shared memory of what your team decided, served to every Claude Code, Codex and Cursor session through a gate you set, Harbor.
What does Supermemory do for coding agents?
Supermemory calls itself a memory and continual learning engine for agents. Its coding plugins key memory to the repository's git remote, so everyone in your Supermemory organization who works on the same repo reads and writes the same memory. In Claude Code, a hook loads the project's profile facts when a session starts, searches for matches before each prompt, and saves the conversation after each reply. The Codex, OpenCode and Cursor plugins write to the same repository container.
It is a strong default for a team: one install per agent, nobody writes anything down, and the core is open source under MIT. Our Harbor vs Supermemory page puts the two side by side, row by row.
Why do teams look for a Supermemory alternative?
These are the limits Supermemory's own pages make visible:
- Nothing is reviewed before it is shared. The Claude Code docs describe captures saved automatically after each reply, with no review step, into a container the whole repo shares.
- It remembers sessions, not decisions. What a teammate said in a session reaches the next one. A decision made in a Slack thread or a pull request review gets there only through a connector or a session that repeats it.
- Seats are capped by plan. The pricing page lists one seat on Free, three on Pro at $19 a month, and unlimited seats from Scale at $399 a month. Usage past each plan's credits is billed.
- Not every agent. Its coding agents page names Claude Code, Cursor, Codex, OpenCode and Hermes. GitHub Copilot is not on it.
What is the best Supermemory alternative for a team of coding agents?
If the team's agents should start from what the team decided, rather than from what past sessions said, Harbor. If you like automatic session memory and want to run it yourself, Cognee. If one developer is the whole team, claude-mem. If a short list of rules covers it, a committed CLAUDE.md or AGENTS.md, and you need no tool at all.
1. Harbor: what the team decided, in every session
Harbor is shared memory for a team's coding agents. It learns what your team decides and corrects, from Slack, GitHub, GitLab, Notion, Confluence, Jira, Linear, Google Drive and Granola, plus what agent sessions record. Hooks add the rules that apply to the session before the agent writes a line, in Claude Code, Cursor, Codex and GitHub Copilot. The same rules are on an MCP server, the CLI and /harbor in Slack.
- Instead of Supermemory because: you set the gate. Manual holds everything for a person, Hybrid adds routine rules and holds the important ones, Auto adds everything else, and secrets, install steps and dangerous commands wait in every mode. Every rule shows how often it was served and how often an answer cited it (which rules got used).
- Also: block rules refuse a risky command before it runs, and when a push deletes the file a rule is about, the rule is retired, or flagged if other repos still use it.
- Watch out for: the service is hosted, with nothing to self-host. Only the CLI, harborloop, is open source, under MIT. It wires four agents, not every one Supermemory does.
- Price: a Free plan, then Pro at $24 per person a month billed yearly, or $30 monthly. Setup is self-serve, with no card and no call.
2. Cognee: open source, with hooks in Claude Code and Codex
An open-source memory platform under Apache-2.0. Its Claude Code plugin looks up context on each prompt through a hook, and it has a Codex plugin too. Cognee says it can connect Slack, GitHub and Linear so agents recall what a company knows.
- Pick it if: you want Supermemory's kind of automatic recall on infrastructure you run.
- Side by side: Harbor vs Cognee.
3. Mem0: one shared pool per repo
A memory layer for AI agents and apps, open source under Apache 2.0. Its Claude Code plugin sends batches to Mem0 after every five exchanges and recalls before the first reply. Everyone on a repo reads and writes the same pool, and personal preferences stay private. The plugin needs a Mem0 Platform API key.
- Pick it if: you are also building an app that has to remember its own users, and want one vendor for both.
- More: Mem0 alternatives for coding agents.
4. Zep: facts that know when they were true
Zep calls itself the enterprise context layer for reliable agents, built on a temporal context graph: each fact records when it was true. Its graph engine, Graphiti, is open source under Apache-2.0. It lists Claude Code, Codex, Cursor, GitHub Copilot and Gemini CLI as coding agents that connect through MCP or its SDKs.
- Pick it if: your agents work on customer or business data where facts go out of date.
- Side by side: Harbor vs Zep.
5. Letta: an agent that owns its memory
Letta Code is a stateful agent harness under Apache-2.0, and all of its context, memory blocks included, is tracked in git. It is an agent of its own, not memory you add to Claude Code or Cursor.
- Pick it if: you want one long-lived agent that learns, more than memory for the agents you already run.
6. ByteRover: memory as files you own
ByteRover keeps memory local first, as markdown files in a context tree under .brv/context-tree/, with cloud sync optional. You choose which files stay private and which are shared with teammates and other agents. Its docs name Cursor, Claude Code and Claude Desktop, among others.
- Pick it if: you want memory you can read, edit and keep in git.
- Side by side: Harbor vs ByteRover.
7. claude-mem: just you, on your machine
Hooks send each tool call to a local worker, a second agent turns it into short observations, and everything stays in a SQLite database in ~/.claude-mem. It is Apache-2.0, with an optional shared TeamBrain where each person chooses what to share. More in our claude-mem note.
When is a plain CLAUDE.md or AGENTS.md enough?
Often. A project CLAUDE.md is shared with your team through version control, and Claude Code loads it at the start of every session. AGENTS.md is an open format that Codex, Cursor, GitHub Copilot's coding agent and many others read. A committed file costs nothing, every change goes through your normal pull request review, and nothing is saved without someone writing it.
It stops being enough when the rules outgrow one file, when the decisions live in Slack and reviews rather than in the repo, or when nobody can tell which lines an agent actually used. See why a growing CLAUDE.md stops working.
How do Supermemory and its alternatives compare?
| Built for | Coding agents | How memory gets in | Shared with the team | Open source or hosting | |
|---|---|---|---|---|---|
| Supermemory | A memory and context engine for agents | Plugins for Claude Code, Cursor, Codex, OpenCode, Hermes | Saved after each reply, no review step | One memory per git remote, across your org | Yes, MIT |
| Harbor | A team's coding agents | Hooks and MCP in Claude Code, Cursor, Codex, Copilot | Learned from Slack, reviews, docs and sessions, through a gate you set | Yes, the team's rules | The CLI, MIT; the service is hosted |
| Cognee | Agent, team and company memory | Plugins for Claude Code and Codex, MCP | A hook on each prompt; Slack, GitHub, Linear | Built for teams, per its own tagline | Yes, Apache-2.0 |
| Mem0 | Memory for agents and apps | Claude Code plugin, MCP | Sent every five exchanges | One pool per repo | The library, Apache 2.0 |
| Zep | Enterprise agents | Claude Code, Codex, Cursor, Copilot, Gemini CLI via MCP or SDKs | Facts in a temporal graph | Access controls per connection | Zep is hosted; its graph engine, Graphiti, is Apache-2.0 |
| Letta | Agents that learn | Letta Code, its own agent | The agent keeps memory blocks in git | Its docs have a shared memory feature | Yes, Apache-2.0 |
| ByteRover | Memory as local files | Cursor, Claude Code and more | Markdown in a context tree | The files you choose | Local first, cloud sync optional |
| claude-mem | One developer | Hooks | Each tool call, summarised | Optional TeamBrain | Yes, Apache-2.0 |
| CLAUDE.md or AGENTS.md | Stable project rules | Claude Code; AGENTS.md in Codex, Cursor, Copilot and more | You write it | Through git and pull requests | A plain file |
Which Supermemory alternative should you pick?
- A few stable rules and a small team: a committed CLAUDE.md or AGENTS.md.
- Supermemory's automatic recall, on your own infrastructure: Cognee.
- Just you, on your machine: claude-mem.
- Facts that change over time, on business data: Zep.
- Memory as files you own and edit: ByteRover.
- One memory of what the team decided, served to every Claude Code, Codex, Cursor and Copilot session through a gate you set, with served and cited counts per rule: Harbor. Sign up, name your workspace, then run
npx harborloop init.
For a ranked list of memory servers a team can share, see the best MCP memory servers for teams. If a line here is out of date, tell us and we will fix it.
Questions
What is the best Supermemory alternative for a team of coding agents?
For one shared memory of what your team decided, served to every Claude Code, Codex, Cursor and Copilot session through a gate you set, Harbor. For automatic session memory you run yourself, Cognee. For one developer, claude-mem. For a few stable rules, a committed CLAUDE.md or AGENTS.md.
Does Supermemory share memory across a team?
Yes. Supermemory's coding plugins key memory to the repository's git remote, so everyone in your Supermemory organization who works on the same repo reads and writes the same memory. Its Claude Code docs describe captures saved after each reply with no review step.
Is there an open-source Supermemory alternative?
Cognee, Mem0, Letta Code and claude-mem are all open source under Apache 2.0, and Zep's graph engine Graphiti is too. Supermemory is itself open source under MIT and can run locally. Harbor's CLI, harborloop, is MIT, but the service is hosted.
Which Supermemory alternative works with Claude Code, Codex and Cursor?
Harbor adds the team's rules through hooks in Claude Code, Cursor, Codex and GitHub Copilot, with an MCP server for the rest. Cognee has Claude Code and Codex plugins. Zep connects Claude Code, Codex, Cursor, Copilot and Gemini CLI through MCP or its SDKs.
Do I need a memory tool, or is CLAUDE.md enough?
A committed CLAUDE.md or AGENTS.md is often enough. It is shared through version control, reviewed in pull requests and free. A tool earns its place when the rules outgrow one file or the decisions live in Slack and reviews rather than in the repo.