claude-multimodel

Switch between model providers in Claude Code — DeepSeek, Gemini, OpenAI, Groq, Kimi, GLM, Qwen, Ollama, Anthropic. Install once, works globally in every project.

Website: https://claude-mm.vatsa.online

Install

npm install -g claude-multimodel

Slash commands register automatically. No config needed.

Usage

Terminal

claude-mm                          # interactive menu → launch Claude
claude-mm launch deepseek          # launch with specific profile
claude-mm list                     # show all profiles + key status
claude-mm active                   # show current profile
claude-mm providers                # list all known providers + compatibility
claude-mm set-key deepseek         # set API key (hidden input)
claude-mm set-model deepseek       # change model for a profile (interactive)
claude-mm set-model deepseek deepseek-r1          # set main model directly
claude-mm set-model deepseek deepseek-r1 deepseek-v4-flash  # set main + subagent
claude-mm add                      # add new profile (guided wizard)
claude-mm remove <profile>         # remove a profile
claude-mm set-default deepseek     # make DeepSeek permanent (writes to shell profile)
claude-mm unset-default            # remove permanent override

Inside Claude Code

/multimodel/switch-profile         show active profile + all options
/multimodel/set-api-key            key setup instructions
/multimodel/add-profile            add profile guidance
/multimodel/help                   all commands

Built-in profiles

| Profile | Provider | Main Model | Subagent | Key needed | Setup | |---------|----------|------------|----------|------------|-------| | claude | Anthropic | (Claude Code default) | (default) | No | None | | claude-max | Anthropic | (Claude Code default, max effort) | (default) | No | None | | deepseek | DeepSeek | deepseek-v4-pro | deepseek-v4-flash | Yes | Direct | | deepseek-flash | DeepSeek | deepseek-v4-flash | deepseek-v4-flash | Yes | Direct | | openrouter | OpenRouter | anthropic/claude-opus-4 | llama-4-maverick | Yes | Direct | | kimi | Moonshot AI | kimi-k2.6 | kimi-k2-turbo | Yes | Direct | | glm | Z.AI (intl) | glm-5.1 | glm-4-flash | Yes | Direct | | glm-cn | Zhipu BigModel | glm-5.1 | glm-4-flash | Yes | Direct (CN) | | openai | OpenAI | gpt-5.5 | gpt-4o-mini | Yes | LiteLLM proxy | | gemini | Google | gemini-3.5-pro | gemini-3.5-flash | Yes | LiteLLM proxy | | groq | Groq | llama-4-70b | llama-3-8b | Yes | LiteLLM proxy |

claude and claude-max set no model overrides — Claude Code always picks its current best models. Zero staleness.

Direct profiles connect to the provider's Anthropic-compatible API without any proxy. LiteLLM proxy profiles require litellm running locally — see setup below.

Set API keys

claude-mm set-key              # interactive menu, hidden input
claude-mm set-key deepseek     # specific profile

Keys stored in ~/.claude/multimodel/profiles.json — local only, never committed.

Native Anthropic profiles read your existing ANTHROPIC_API_KEY. No setup needed.

Change model for a profile

claude-mm set-model deepseek                              # interactive
claude-mm set-model deepseek deepseek-r1                  # set main model
claude-mm set-model deepseek deepseek-r1 deepseek-v4-flash  # main + subagent

Shows current models, hints popular models for that provider, then saves.

Updates: ANTHROPIC_MODEL, ANTHROPIC_DEFAULT_OPUS_MODEL, ANTHROPIC_DEFAULT_SONNET_MODEL, ANTHROPIC_DEFAULT_HAIKU_MODEL, CLAUDE_CODE_SUBAGENT_MODEL.

Or edit directly:

# open profiles file
code ~/.claude/multimodel/profiles.json    # VS Code
notepad $env:USERPROFILE\.claude\multimodel\profiles.json  # Windows

Add a custom model or provider

claude-mm add

Guided wizard:

1. Pick from known providers list (or "manual" for custom endpoint)
2. Enter profile name
3. Enter model ID (flagship + fast/subagent)
4. Set API key

Custom model from existing provider — pick the provider, enter any model ID they offer:

Provider: deepseek
Profile name: deepseek-reasoner
Main model: deepseek-reasoner
Subagent: deepseek-v4-flash

New provider — pick "custom", enter base URL + model IDs:

Provider: custom
Base URL: https://api.myprovider.com/anthropic
Main model: my-model-v1

Known providers

View the full registry:

claude-mm providers

| Provider | Key | Compatibility | Setup | |----------|-----|---------------|-------| | Anthropic | anthropic | Native | None | | DeepSeek | deepseek | Anthropic-compatible | Direct | | OpenRouter | openrouter | Anthropic-compatible | Direct | | Kimi (Moonshot) | kimi | Anthropic-compatible | Direct | | GLM / Z.AI (intl) | glm | Anthropic-compatible | Direct | | GLM / Zhipu (CN) | glm-cn | Anthropic-compatible | Direct (CN) | | OpenAI | openai | OpenAI-only | LiteLLM proxy | | Google Gemini | gemini | OpenAI-only | LiteLLM proxy | | Groq | groq | OpenAI-only | LiteLLM proxy | | Mistral AI | mistral | OpenAI-only | LiteLLM proxy | | Together AI | together | OpenAI-only | LiteLLM proxy | | Qwen (Alibaba) | qwen | OpenAI-only | LiteLLM proxy | | Ollama (local) | ollama | OpenAI-only | LiteLLM proxy | | Custom | custom | Anthropic-compatible | Direct |

Why LiteLLM for some? Claude Code sends requests in Anthropic's /v1/messages format. OpenAI, Groq, Gemini etc. only speak /v1/chat/completions — a different wire protocol. LiteLLM translates between them locally. DeepSeek, OpenRouter, Kimi, and GLM all ship real Anthropic-compatible endpoints so they connect directly with no proxy.

Providers that need LiteLLM proxy

Install LiteLLM once:

pip install 'litellm[proxy]'

Start the proxy for your provider (in a separate terminal, before launching Claude):

# OpenAI
litellm --model openai/gpt-4.1 --api_key $OPENAI_API_KEY --port 4000

# Google Gemini (get key at aistudio.google.com)
litellm --model gemini/gemini-2.5-pro --api_key $GEMINI_API_KEY --port 4000

# Groq (Llama 4 Maverick)
litellm --model groq/llama-4-maverick-17b-128e-instruct --api_key $GROQ_API_KEY --port 4000

# Mistral
litellm --model mistral/mistral-large-latest --api_key $MISTRAL_API_KEY --port 4000

# Together AI
litellm --model together_ai/meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8 --api_key $TOGETHERAI_API_KEY --port 4000

# Ollama (local, no API key)
litellm --model ollama/llama3.3 --api_base http://localhost:11434 --port 4000

Then run claude-mm launch <profile>. The CLI will remind you with the exact command if LiteLLM isn't set up.

OpenRouter, Kimi, and GLM connect directly — no LiteLLM needed. See the Direct providers section.

OpenRouter popular models (set via claude-mm set-model openrouter <model-id>):

anthropic/claude-opus-4
openai/gpt-4.1
openai/o3
google/gemini-2.5-pro
deepseek/deepseek-chat
deepseek/deepseek-r1
meta-llama/llama-4-maverick
moonshotai/kimi-k2.6
x-ai/grok-3

Make a provider permanent

Want DeepSeek every time — even with plain claude?

claude-mm set-default deepseek
# restart terminal
claude                          # now uses DeepSeek

Writes env vars into your shell profile ($PROFILE on Windows, .bashrc/.zshrc on Mac/Linux).

Revert:

claude-mm unset-default
# restart terminal

How it works

• Profiles = named sets of env vars
• claude-mm launch sets env vars in child process → runs claude
• Env vars must be set at launch — switching mid-session requires relaunch
• set-default writes vars to shell profile so every terminal already has them
• Slash commands installed to ~/.claude/commands/multimodel/ — available in every project

Env vars per profile

| Var | Purpose | |-----|---------| | ANTHROPIC_BASE_URL | API endpoint (empty = native Anthropic) | | ANTHROPIC_AUTH_TOKEN | API key for third-party providers | | ANTHROPIC_MODEL | Default model | | ANTHROPIC_DEFAULT_OPUS_MODEL | Model when Opus is requested | | ANTHROPIC_DEFAULT_SONNET_MODEL | Model when Sonnet is requested | | ANTHROPIC_DEFAULT_HAIKU_MODEL | Model when Haiku is requested | | CLAUDE_CODE_SUBAGENT_MODEL | Model for background subagents | | CLAUDE_CODE_EFFORT_LEVEL | low / medium / high / max |

Update

npm update -g claude-multimodel