claude-multi
Back to use cases

For indie developers

Ship side projects faster by running multiple AI providers side-by-side. One terminal, zero config clashes.

What you're dealing with

  • Swapping API keys in environment variables every time you switch models
  • One ~/.claude directory means one model at a time, or manual config juggling
  • Testing how different models handle the same prompt means starting over each time
  • Plugins and skills you installed for Claude break when you point it at another provider
  • No way to keep a cheap model for boilerplate and a strong model for architecture in parallel

What changes

  • Run claude-deepseek for boilerplate and claude-anthropic for hard problems at the same time
  • Each instance keeps its own history, settings, and plugins — no cross-contamination
  • Plugin sync means you install a skill once and every instance picks it up
  • One command to add a new provider, one command to launch it
  • MIT licensed, runs locally, no accounts to create

Recommended providers

DeepSeek GLM Qwen

You build side projects. You don’t have time to babysit your tooling.

The standard workflow is: install Claude Code, point it at Anthropic, ship. That works until you realize DeepSeek is cheaper for boilerplate, or GLM handles your codebase better, or you want to A/B test prompts across models. Then you’re editing env vars, copying config files, and praying nothing breaks.

claude-multi gives each model its own directory. claude-deepseek for the grunt work. claude-anthropic when you need the heavy lifting. Both running at the same time, both with their own history and settings, neither one touching the other.

How it fits into your day

You’re working on a SaaS app. The backend is mostly CRUD — DeepSeek or Qwen handles that fine at a fraction of the cost. The auth module is tricky — switch to Anthropic for that part. Both instances stay open. Both keep their context. No switching cost.

When you install a new plugin or skill, symlinks carry it to every instance. Update once, done.

The setup

Terminal window
bun add -g claude-multi
claude-multi
# Pick "Add new instance", choose a provider, paste your key
# Done. Run `claude-deepseek` or whatever alias it gives you.

Two minutes. No config files to write. No daemons running in the background.

What you’re probably comparing this to

  • Multiple Claude Code installs: Manually creating ~/.claude-deepseek, writing wrapper scripts, managing symlinks yourself. claude-multi automates all of it.
  • A proxy router: claude-multi isn’t a proxy. Each instance is a real, independent Claude Code environment. You get the full CLI, all flags, all plugins.

Where to go next