Cross-Agent Memory Sharing

Brain Memory uses a single global directory — ~/.brain/ — that is shared across all supported AI agents. A decision stored by Claude Code in one project is immediately available to Gemini CLI, OpenAI Codex CLI, or OpenCode in any other project, with no configuration, no export, and no per-project setup.

Model-Agnostic, Too

Portability runs along two independent axes:

Agent-portable — the four supported agents (Claude Code, Gemini CLI, OpenAI Codex CLI, OpenCode) all read and write the same brain.
Model-agnostic — because memory is plain files, not embeddings welded to a particular model, the LLM driving your agent can be anything (GPT, Claude, Gemini, or any model routed through a gateway). Switching the underlying model changes nothing about your memory.

Put together: switch your model or switch your agent, and you keep remembering — pick up exactly where you left off.

How It Works

All four supported agents read and write to the same ~/.brain/ directory:

~/.brain/                          ← Single shared directory
├── index.json                     ← Shared index
├── associations.json              ← Shared associative network
├── contexts.json                  ← Session contexts from all agents
├── professional/
│   └── decisions/
│       └── chose-kafka.md         ← Created by Claude Code
├── personal/
│   └── education/
│       └── rust-ownership.md      ← Created by Gemini CLI
├── professional/
│   └── learnings/
│       └── async-pitfalls.md      ← Created by OpenCode
└── ...

The format is agent-agnostic: plain Markdown files with YAML frontmatter. Any tool that can read Markdown can read your brain.

Agent-Agnostic Format

Brain Memory deliberately avoids any agent-specific constructs:

Component	Format	Why
Memory files	Markdown with YAML frontmatter	Universal, human-readable
Index	JSON	Supported by every language and tool
Associations	JSON	Simple weighted edge graph
Configuration	JSON (in index.json)	No proprietary config format
Directory structure	Plain filesystem	Works on macOS, Linux, Windows

No vendor lock-in. No proprietary binary formats. Your memories are yours.

What Gets Shared

Everything in ~/.brain/ is automatically shared:

Memories — All memory files, regardless of which agent created them
Associations — The full associative network, including links created by different agents
Index — The central registry with all memory metadata
Configuration — Brain settings apply to all agents uniformly
Review queue — Spaced repetition schedule works across agents
Contexts — Session context history from all agents (labeled by source)

What Does Not Get Shared

Agent configuration — Each agent's command prompts and prompt sections are installed separately in their own config directories (~/.claude/, ~/.gemini/, ~/.codex/, ~/.config/opencode/)
Session state — Each agent maintains its own session context during active use
Sync state — The ~/.brain/.sync/ directory is local and agent-independent

Cross-Agent Scenarios

Scenario 1: Claude Code creates, Gemini CLI recalls

You work on a project with Claude Code and run /brain:memorize
Claude Code stores a decision memory in ~/.brain/professional/
Later, you open a different project with Gemini CLI
Gemini CLI loads the brain at session start and sees the memory
When the topic comes up, Gemini CLI references it naturally

Scenario 2: Multiple agents in the same session

You use Claude Code to analyze a codebase and memorize findings
You switch to Gemini CLI for a different task in the same project
Both agents share the same brain — Gemini sees what Claude stored
Gemini CLI adds its own memories, which Claude Code will see next time

Scenario 3: Review queue across agents

/brain:sleep is run from Claude Code, populating the review queue
You open Gemini CLI and run /brain:sleep
Its replay pass uses the same queue generated by Claude Code
Recalled memories are reinforced regardless of which agent does the review

No Configuration Needed

Cross-agent sharing requires no setup beyond installing Brain Memory for each agent. Since all agents read from and write to ~/.brain/, sharing is automatic.

The only requirement is that each agent has its own command prompts installed. During installation, you can choose to install for multiple agents at once:

npm install -g brain-memory@beta
brain --all --global

This installs the CLI tools and configures command prompts for Claude Code, Gemini CLI, OpenAI Codex CLI, and OpenCode simultaneously, all pointing to the same ~/.brain/ directory.

info

Each agent has slightly different command syntax (slash commands for Claude/Gemini/OpenCode, skills for Codex), but they all operate on the same underlying brain structure and data format.

tip

Identical, deterministic recall across agents relies on the brain CLI being on your PATH — always install with npm install -g (not npx), so every agent shells out to the same scoring engine instead of falling back to manual file operations.

tip

If you primarily use one agent but occasionally switch to another, your brain context follows you seamlessly. There is no migration or export step — just start using the other agent.

Human Access

Because the format is plain Markdown, you can also browse and edit your brain directly:

Open ~/.brain/ in VS Code, Obsidian, or any file explorer
Read memory files with any text editor
Search across memories with grep or your editor's search
Edit memories manually if needed (be careful with frontmatter format)
Use Git to track changes: cd ~/.brain && git init && git add -A && git commit -m "snapshot"

Your brain is not locked inside any tool — it is always accessible.