claude-multi
Back to FAQ
Architecture

Is claude-multi a fork of Claude Code?

No. It doesn’t fork, patch, or modify Claude Code in any way.

The actual mechanism

Each instance is a shell wrapper script (batch file on Windows) that does two things: set CLAUDE_CONFIG_DIR to the instance’s directory, then exec the real claude binary. Simplified:

#!/bin/sh
export CLAUDE_CONFIG_DIR="$HOME/.claude-multi/deepseek"
exec claude "$@"

That’s it. No proxy, no monkey-patching, no background process. Claude Code reads its config from the pointed-to directory instead of ~/.claude, and everything else, flags, commands, keybindings, works exactly as it does normally.

Why this matters in practice

When Claude Code ships an update, you get it immediately. There’s nothing to rebase or merge. Every feature works because you’re running the actual binary. And each instance is a real directory you can cd into, inspect, or delete with standard tools.

What’s inside an instance directory

~/.claude-multi/deepseek/
├── settings.json        # provider env vars + merged settings
├── .claude.json         # instance-level Claude config
├── plugins/             # symlinked or copied plugins
├── skills/              # symlinked or copied skills
└── projects/            # conversation history per project

More info

  • /about/: the “wrapper, not a fork” explanation
  • /docs/how-it-works/: architecture overview
  • Run claude-multi, pick an instance, select Instance details to see its directory, wrapper path, and config
  • src/wrapper.ts: wrapper generation code
  • src/config.ts: how instance directories are created