One harness for
every Claude
you run.
Run Claude Code against Anthropic, GLM, MiniMax, DeepSeek, MiMo, Kimi, Qwen and more from the same machine. Each one keeps its own config, plugins, and history. Nothing overlaps. One terminal UI runs the whole thing.
🤖 Claude Multi Interactive Mode
▸ ➕ Add new instance
📋 List all instances
🔌 Manage plugins
🔄 Sync mode
⚙️ MCP servers
🚪 Exit
One command. Any runtime.
It's on npm. Install it globally with whatever package manager you already use. No curl-pipe-sh, no platform binaries, no funny business.
For when one model isn't enough.
Stop trampling your .claudedirectory every time you swap models.
Many providers, one harness
Templates for GLM, MiniMax, DeepSeek, MiMo, Kimi, Qwen and more. Every provider you set up gets its own alias (claude-kimi, claude-qwen) that drops straight into your terminal.
Isolated configs
Every instance has its own settings.json, history, and credentials. Your Anthropic setup never touches your GLM setup.
Plugin sync
Plugins and skills can symlink back to your primary install. Install once, every instance picks it up.
Real Claude underneath
Wraps the official claude binary. Every flag, every slash command, every plugin, exactly as you know them.
Local credentials
API keys live in per-instance config files. No global env vars leaking into the wrong session.
Ready for every model.
Endpoints, model names, defaults. Already wired up. Pick a provider in the TUI, paste your API key, you're done.
Anthropic
Claude Opus 4.7, Sonnet 4.6, Haiku 4.5
api.anthropic.com
GLM Coding Plan
GLM-5.3, GLM-5-Turbo
api.z.ai
MiniMax
MiniMax-M3
api.minimax.io
DeepSeek
V4-Pro, V4-Flash
api.deepseek.com
Xiaomi MiMo
MiMo-V2.5-Pro, MiMo-V2.5
api.xiaomimimo.com
Moonshot Kimi
Kimi K2.7 Code, K2.6, K2.5
api.moonshot.ai
Alibaba Qwen
Qwen3-Coder-Next, Plus, Flash
dashscope-intl.aliyuncs.com
Two steps. No flags to remember.
- 1
Launch the TUI
Type
claude-multiand hit enter. The menu lets you add an instance, pick a provider, paste your key, choose what to copy from your default Claude setup, and you're done. Arrow keys to move, enter to select, that's it. - 2
Run your new alias
The wizard prints something like
claude-glm. Run it. Every flag and slash command fromclaudeworks exactly as you'd expect.
$claude-multi
▸ ➕ Add new instance
Step 1 / 8 · Instance Name › glm
Step 2 / 8 · Provider › GLM
Step 3 / 8 · API Key › ••••••••••••
✓ Instance 'glm' created.
$claude-glm
The whole flow lives in the menu. No subcommands, no flags, nothing to memorize.
The TUI does everything.
Type claude-multi and the terminal UI opens. Add, list, manage plugins, swap MCP servers, repair symlinks. No flags to remember, ever.
One tool. Every developer.
Different people run claude-multi for different reasons.
Freelancers
Keep client work strictly separated. Different projects, different models, different configs — all from one machine.
Indie developers
Ship side projects faster by running multiple AI providers side-by-side. One terminal, zero config clashes.
Researchers
Run the same prompt against multiple models, compare outputs side-by-side, keep every experiment isolated and reproducible.
Students
Learn AI-assisted coding with free and cheap models. No credit card required for most providers, no complex setup.
Teams
Standardize how your team uses AI coding tools. Shared provider configs, consistent setups, no "works on my machine" incidents.
Run every Claude.
Conflict-free.
Open source. MIT licensed. Pick whichever model fits the job.