---
title: "Claude Code memory for teams: share what agents learn"
description: "Claude Code memory has two parts: CLAUDE.md files you commit and auto memory that stays on one machine. How teams share both, and where it stops."
image: "https://bombos.ai/assets/og-en-v2.png"
---

# Claude Code memory for teams: how do you share what your agents learn?

Claude Code memory has two parts: CLAUDE.md files you commit and auto memory that stays on one machine. How teams share both, and where it stops.

*Piet Baudoin, Bombos · published November 16, 2026 · November 16, 2026 · updated November 16, 2026 · November 16, 2026 · 8 min read*

**Does your AI run into this too?** Together we map what your AI forgets. Free, and the plan is yours, with or without Bombos.

[Book your half hour](https://bombos.ai/en/contact/)

**Claude Code memory has two parts: CLAUDE.md files that you write and can commit to the repo, and auto memory that Claude writes itself and that Anthropic describes as “machine-local”. A team shares the first through git and loses the second: what one engineer’s agent learned on Monday is not on a colleague’s laptop on Tuesday. To share it, promote lessons into a committed CLAUDE.md or `.claude/rules/`, enforce what must never be skipped with hooks, and put knowledge that spans tools in a memory layer outside the repo.**

By Bombos. Published November 16, 2026. Sources checked November 16, 2026.

## The short answer

-   **CLAUDE.md is the shared part.** Project instructions in `./CLAUDE.md` or `./.claude/CLAUDE.md` reach “Team members via source control” ([Claude Code docs, memory](https://code.claude.com/docs/en/memory)).
-   **Auto memory is per machine.** “Auto memory is machine-local… Files are not shared across machines or cloud environments” (same source).
-   **Both are context, not rules.** “Claude treats them as context, not enforced configuration.” For something that must always happen, use a hook (same source).
-   **Organisations can push one CLAUDE.md** to every machine through managed settings or MDM.
-   **Nothing here crosses tools.** Cursor, Codex and ChatGPT read their own files; AGENTS.md is the closest thing to a shared format.

## How does Claude Code memory work?

Claude Code loads instructions at the start of each session from several places, in this order, from broad to narrow; later files appear later in context and are concatenated, not overridden ([Claude Code docs, memory](https://code.claude.com/docs/en/memory)):

| Location | Who writes it | Who gets it |
| --- | --- | --- |
| Managed policy CLAUDE.md | IT, via MDM or the `claudeMd` key in managed-settings.json | Everyone in the organisation |
| User `~/.claude/CLAUDE.md` | You | You, in every project |
| Project `./CLAUDE.md` or `./.claude/CLAUDE.md` | The team | Team members via source control |
| `.claude/rules/*.md` | The team | Loaded for matching paths |
| `CLAUDE.local.md` | You | You, in this project only |
| Auto memory | Claude | You, on this machine |

A CLAUDE.md can import other files with `@path`, up to four levels deep. Anthropic advises keeping each file under 200 lines. By default, Claude Code reads AGENTS.md instead only when there is no `CLAUDE.md`, `.claude/CLAUDE.md` or `CLAUDE.local.md` in the working directory or any directory above it (from version 2.1.277). The “Project instructions” setting `claude-md-and-agents-md` loads both ([Claude Code docs, AGENTS.md](https://code.claude.com/docs/en/memory#agentsmd)).

**Auto memory** is what Claude writes on its own: notes about you, feedback you gave, the project, and references, stored as files under `~/.claude/projects/<project>/memory/` with a `MEMORY.md` index. Each session loads the first 200 lines or 25 KB of that index. You can move it with the `autoMemoryDirectory` setting.

## What does a team lose with machine-local memory?

It loses the corrections. Say an engineer tells Claude Code: “never run migrations against staging from a branch, only from main.” Auto memory may save that as feedback. The engineer next to them never sees it, and neither does the CI agent running in the cloud. The lesson exists once, on one laptop.

Three things follow:

1.  **The same mistake gets corrected several times.** Each engineer teaches their own agent.
2.  **Knowledge leaves with the laptop.** When someone changes teams or leaves, their auto memory goes with their machine.
3.  **Nobody reviews it.** What sits in auto memory has not passed code review, so it can also hold a wrong lesson that nobody else can see.

A public feature request, “Shared Team Memory for Claude Code” (issue 38536, opened March 25, 2026, mirrored on claudeissues.com), shows teams run into this. Anthropic’s documentation does not describe a shared auto memory today.

## How do you share Claude Code memory across a team?

There are four workable patterns, from simple to structural:

1.  **Promote lessons into the committed CLAUDE.md.** Make it a habit, for example in the pull request template: “Did you correct the agent this week? Add the rule.” The rule then goes through review like code.
2.  **Split by path with `.claude/rules/`.** Rules that apply only to `migrations/` or `frontend/` load only there, which keeps the main file short. Symlinks let several repositories share one rules folder; rules linked from outside the repo load only after you approve external imports, and path-scoped ones among them do not load.
3.  **Point `autoMemoryDirectory` at a shared location.** Technically possible; in practice you then share unreviewed notes, including personal ones, and you have to handle conflicts yourself. We would only do this with a review step.
4.  **Enforce with hooks, not instructions.** If a migration must never run from a branch, a PreToolUse hook that blocks the command is reliable. A sentence in CLAUDE.md is context the model weighs, not a guarantee.

For an organisation-wide baseline (security rules, approved tools), use the managed policy CLAUDE.md, which IT deploys to every machine.

## What about Cursor, Codex and the rest of the team?

Most teams do not run one agent. Cursor reads Project Rules from `.cursor/rules` and Team Rules: “Team and Enterprise plans can create and enforce rules across their entire organization” ([Cursor docs, rules](https://cursor.com/docs/context/rules)). Codex and GitHub Copilot read AGENTS.md, and Gemini CLI can be configured to. The Linux Foundation says AGENTS.md is used by “more than 60,000 open source projects and agent frameworks” ([Linux Foundation, December 9, 2025](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation)).

| Aspect | Claude Code | Cursor | Codex and others |
| --- | --- | --- | --- |
| Shared file in the repo | CLAUDE.md, `.claude/rules/` | `.cursor/rules`, AGENTS.md | AGENTS.md |
| Org-wide rules | Managed policy CLAUDE.md | Team Rules (Team, Enterprise) | Varies |
| Memory the agent writes | Auto memory, machine-local | Not covered in the sources we read | Varies |
| Reads company knowledge outside the repo | Via MCP servers | Via MCP servers | Via MCP servers |

A practical bridge: keep the shared rules in AGENTS.md. Claude Code reads it directly when there is no `CLAUDE.md`, `.claude/CLAUDE.md` or `CLAUDE.local.md` in the working directory or above it. A personal `CLAUDE.local.md` is enough to stop that, so if anyone keeps one, add `@AGENTS.md` at the top of the CLAUDE.md, or set Project instructions to `claude-md-and-agents-md`. Then one file feeds every coding agent. It still does not cover the product manager in ChatGPT or the support lead in Copilot; see [ChatGPT Team or Copilot: what does it remember of your team](https://bombos.ai/kennisbank/chatgpt-team-of-copilot-wat-onthoudt-het-van-je-team/) (Dutch).

## When do you need a memory layer instead of files?

Files in the repo work for rules about the code. They stop working when:

-   the knowledge is not about one repository (customers, pricing, decisions from a call);
-   the same knowledge must reach non-coding tools;
-   you want corrections to become rules without someone editing a file by hand;
-   you need to see which source an answer came from.

That is the problem memory layers try to solve. Mem0, Zep, Letta and Supermemory are building blocks for developers who build their own agents; Supermemory also ships plugins for Claude Code, Cursor and Codex. We compare them in [Mem0 alternatives for team memory](https://bombos.ai/en/knowledge-base/mem0-alternatives-for-team-memory/). The wider discipline, deciding what an agent sees at each step, is [context engineering](https://bombos.ai/en/knowledge-base/what-is-context-engineering/).

## How does Bombos do this?

Bombos is the memory layer under the AI tools a team already uses: Claude Code, Claude, Codex, Cursor, ChatGPT, Copilot, or your own agents. You learn from AI. But AI doesn’t learn from you. With Bombos it does.

-   **Learned rules.** A correction from one engineer becomes a rule about how the team works, applied before anyone asks. Nobody has to remember to update CLAUDE.md.
-   **One memory for every agent.** The same rules and knowledge in Claude Code, Cursor and ChatGPT, through an open connection such as MCP.
-   **It reads what already exists.** Mail, documents, business systems and repositories, not just what someone typed into a file.
-   **Traceable.** A context graph with retrieval (similarity, keywords, proximity in the graph, rules that apply only in a given context), not a model of its own. You can see what an answer rests on.
-   **Cloud or self-hosted**, on your own server with your own models if you want.

Bombos is young; in the call we tell you which connections work with your tools today. One person can always start: [try the plugin](https://bombos.ai/en/waitlist/) for your own agents.

## Frequently asked questions

**Does Claude Code share memory between team members?** CLAUDE.md files in the repository are shared through git. Auto memory is machine-local and is not shared.

**Where is Claude Code auto memory stored?** In `~/.claude/projects/<project>/memory/`, with a `MEMORY.md` index. The location can be changed with `autoMemoryDirectory`.

**Should CLAUDE.md or AGENTS.md be the source of truth?** If your team uses more than one coding agent, make AGENTS.md the shared file. By default Claude Code reads it only when there is no `CLAUDE.md`, `.claude/CLAUDE.md` or `CLAUDE.local.md` in the working directory or above it; otherwise import it with `@AGENTS.md` or switch the setting to `claude-md-and-agents-md`.

**Can Claude Code be forced to follow a rule?** Not through memory. Anthropic says memory is context, not enforced configuration. Use hooks or permissions for hard rules.

[Book your half hour](https://bombos.ai/en/contact/): together we map what your AI forgets. Free, and the plan is yours, with or without Bombos.

## Which sources did we use?

All sources opened and read on November 16, 2026.

1.  **Claude Code docs, How Claude remembers your project**: [https://code.claude.com/docs/en/memory](https://code.claude.com/docs/en/memory)
2.  **Cursor docs, Rules**: [https://cursor.com/docs/context/rules](https://cursor.com/docs/context/rules)
3.  **Linux Foundation, Agentic AI Foundation announcement** (December 9, 2025): [https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation](https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation)
4.  **AGENTS.md**: [https://agents.md](https://agents.md/)

## What does your AI forget?

1.  01You tell us which AI you use and where it keeps starting over.
2.  02We show what a shared memory changes there, on your own examples.
3.  03You keep the plan, with or without Bombos.

![Piet Baudoin](https://bombos.ai/assets/team/piet-klein.jpg)

**Piet Baudoin**You talk to Piet, from the Bombos teamDutch team from Utrecht and Nijmegen · data in Europe

## More about memory and context

-   [Shared memory across coding agents: Claude Code, Codex and Cursor](https://bombos.ai/en/knowledge-base/shared-memory-across-coding-agents/)
-   [Memory layer for AI agents: what it does, where it sits, and build or buy](https://bombos.ai/en/knowledge-base/memory-layer-for-ai-agents/)
-   [Best MCP memory server for a team: what to compare and which options exist](https://bombos.ai/en/knowledge-base/best-mcp-memory-server-for-a-team/)
-   [Glean alternative for startups and small teams: search, connectors or a memory layer?](https://bombos.ai/en/knowledge-base/glean-alternative-for-startups-and-small-teams/)
-   [How do you make AI agents learn from corrections?](https://bombos.ai/en/knowledge-base/how-to-make-ai-agents-learn-from-corrections/)

[All questions in the knowledge base →](https://bombos.ai/en/knowledge-base/)


---

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://bombos.ai/#organization","name":"Bombos","url":"https://bombos.ai/","logo":"https://bombos.ai/assets/bombos-bee.png","description":"Bombos is the memory layer under every AI your team uses. One shared memory for Claude, ChatGPT, Copilot, Cursor and your own agents, that learns how you work.","sameAs":["https://www.linkedin.com/company/bombos-ai","https://www.youtube.com/@truebombos","https://www.instagram.com/truebombos/","https://bsky.app/profile/bombostrue.bsky.social","https://www.threads.com/@truebombos","https://mastodon.social/@Bombos","https://www.tiktok.com/@truebombos","https://t.me/truebombos","https://www.pinterest.com/truebombos/","https://github.com/bombos-ai","https://linktr.ee/bombos.ai","https://bombosai.substack.com","https://hashnode.com/@bombosai","https://hackernoon.com/u/bombos","https://dev.to/bombos"],"email":"info@bombos.ai","areaServed":["NL","Europe"],"knowsLanguage":["nl","en"],"contactPoint":{"@type":"ContactPoint","email":"info@bombos.ai","contactType":"sales","url":"https://bombos.ai/en/contact/","availableLanguage":["nl","en"]}},{"@type":"WebSite","@id":"https://bombos.ai/#website","name":"Bombos","url":"https://bombos.ai/","inLanguage":["nl-NL","en"],"publisher":{"@id":"https://bombos.ai/#organization"}},{"@type":"Person","@id":"https://bombos.ai/#piet","name":"Piet Baudoin","url":"https://bombos.ai/over-ons/","worksFor":{"@id":"https://bombos.ai/#organization"},"sameAs":["https://www.linkedin.com/in/piet-baudoin","https://x.com/marketeerpiet","https://dev.to/marketeerpiet","https://medium.com/@piet_7573"]},{"@type":"Article","@id":"https://bombos.ai/en/knowledge-base/claude-code-memory-for-teams/#article","headline":"Claude Code memory for teams: how do you share what your agents learn?","description":"Claude Code memory has two parts: CLAUDE.md files you commit and auto memory that stays on one machine. How teams share both, and where it stops.","datePublished":"2026-11-16T00:00:00.000Z","dateModified":"2026-11-16T00:00:00.000Z","author":{"@id":"https://bombos.ai/#piet"},"publisher":{"@id":"https://bombos.ai/#organization"},"image":"https://bombos.ai/assets/og-en-v2.png","mainEntityOfPage":"https://bombos.ai/en/knowledge-base/claude-code-memory-for-teams/","isPartOf":{"@id":"https://bombos.ai/#website"},"inLanguage":"en"},{"@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Bombos","item":"https://bombos.ai/en/"},{"@type":"ListItem","position":2,"name":"Knowledge base","item":"https://bombos.ai/en/knowledge-base/"},{"@type":"ListItem","position":3,"name":"Memory and context","item":"https://bombos.ai/en/knowledge-base/#memory"},{"@type":"ListItem","position":4,"name":"Claude Code memory for teams: how do you share what your agents learn?","item":"https://bombos.ai/en/knowledge-base/claude-code-memory-for-teams/"}]},{"@type":"FAQPage","@id":"https://bombos.ai/en/knowledge-base/claude-code-memory-for-teams/#faq","mainEntity":[{"@type":"Question","name":"How does Claude Code memory work?","acceptedAnswer":{"@type":"Answer","text":"Claude Code loads instructions at the start of each session from several places, in this order, from broad to narrow; later files appear later in context and are concatenated, not overridden (Claude Code docs, memory):"}},{"@type":"Question","name":"What does a team lose with machine-local memory?","acceptedAnswer":{"@type":"Answer","text":"It loses the corrections. Say an engineer tells Claude Code: \"never run migrations against staging from a branch, only from main.\" Auto memory may save that as feedback. The engineer next to them never sees it, and neither does the CI agent running in the cloud. The lesson exists once, on one laptop."}},{"@type":"Question","name":"How do you share Claude Code memory across a team?","acceptedAnswer":{"@type":"Answer","text":"There are four workable patterns, from simple to structural:"}},{"@type":"Question","name":"What about Cursor, Codex and the rest of the team?","acceptedAnswer":{"@type":"Answer","text":"Most teams do not run one agent. Cursor reads Project Rules from `.cursor/rules` and Team Rules: \"Team and Enterprise plans can create and enforce rules across their entire organization\" (Cursor docs, rules). Codex and GitHub Copilot read AGENTS.md, and Gemini CLI can be configured to. The Linux Foundation says AGENTS.md is used by \"more than 60,000 open source projects and agent frameworks\" (Linux Foundation, December 9, 2025)."}}]}]}
```
