Shared memory for Claude Code, across sessions, machines and your team.
Claude Code remembers through CLAUDE.md files and auto memory, and auto memory is machine-local. Levirge Brain gives Claude Code a shared memory over MCP: what one agent learns, every session, machine and teammate can search, with citations back to the source. Install takes two commands.
CLAUDE.md, auto memory, and what neither shares
“Each Claude Code session begins with a fresh context window,” and “Auto memory is machine-local.” Anthropic’s Claude Code memory docs, checked 2026-11-07.
| CLAUDE.md | Auto memory | Levirge Brain | |
|---|---|---|---|
| Who writes it | You | Claude | Your agents, and you |
| What it holds | Instructions and rules | Learnings and patterns | Decisions, findings, fixes, with sources |
| Where it lives | In the repo or ~/.claude | ~/.claude/projects/<project>/memory/ | One Vault per project, on Brain |
| Shared with | Teammates, via git | This machine only | Every machine, teammate and MCP client |
| How you find things | Loaded every session | First 200 lines or 25 KB loaded | search_knowledge, cited answers |
Two commands, then sign in
Restart Claude Code and sign in when the browser opens. Verify with kb_overview. Step-by-step →
| claude plugin marketplace add levirge/brain |
| claude plugin install brain@levirge-brain |
What one agent learns, every agent can use
One limit: an agent only remembers what it captures. Install the plugin, or tell the agent to call add_knowledge.
One Vault per project
A whole team shares one; unrelated projects stay out of each other’s answers.
Cold sessions start warm
/brain:start opens a work session with the overview, context and inbox already loaded.
Cited answers
Captures are distilled into summaries that cite their sources and rebuild when a source changes. /brain:lineage traces a fact.
Handoffs between agents
/brain:handoff sends work, with its context, to another agent or a CI worker.
Per-user tokens
Issue and revoke them; limit them to chosen Vaults or make them read-only.
Not just Claude Code
Codex, Cursor and other MCP clients read the same Vault. People can edit it in Obsidian with two-way sync. Across tools →
Claude Code memory, answered
Does Claude Code remember things between sessions?
Yes, in two ways. CLAUDE.md files hold instructions you write, and auto memory holds notes Claude writes itself. Both load at the start of every session. Auto memory is kept per repository and is machine-local, so it is not shared across machines or cloud environments.
How do I share Claude Code memory with my team?
Share standing instructions by committing a project CLAUDE.md to git. For what agents learn while they work, connect Levirge Brain: every teammate's Claude Code reads and writes the same Vault over MCP, on any machine.
Does Levirge Brain replace CLAUDE.md?
No. Keep CLAUDE.md for rules and conventions you write. Brain holds what agents capture as they work, such as decisions, findings and fixes, and answers questions about it with citations to the source documents.
Does it work with agents other than Claude Code?
Yes. Brain is an MCP server, so Claude Desktop, Codex, Cursor, VS Code, Gemini CLI, Windsurf, Cline and other MCP clients read the same knowledge base.
Can I limit what an agent can see or change?
Yes. Tokens can be limited to chosen Vaults and made read-only. Personal Vaults are private to their owner and fail closed.
What does it cost?
Personal is $20 a month and Personal Pro $60 a month, both for one user. Teams is $35 per seat a month with a minimum of five seats. Every plan includes unlimited agents and unlimited Vaults.
Give every Claude Code session your team's memory
Unlimited agents and Vaults on every plan. Personal from $20 a month; Teams $35 per seat.