provider-switcher

name	provider-switcher
description	Set up and use the hermes-provider-switcher plugin to run Claude Code with alternative LLM providers (GLM, Kimi, MiniMax).
version	1.1.0
author	roach
license	MIT
metadata	{"hermes":{"tags":["Provider","Claude-Code","GLM","Kimi","MiniMax","Plugin"],"related_skills":["claude-code","hermes-plugin-dev","provider-switcher-setup"]}}

Provider Switcher Setup

Install and configure the hermes-provider-switcher plugin to run Claude Code with alternative LLM backends.

Prerequisites

• Hermes Agent installed
• Claude Code CLI: npm install -g @anthropic-ai/claude-code
• gh CLI (optional, for starring the repo)
• API keys for desired providers (set as environment variables)

Installation

Step 1: Clone and install

git clone https://github.com/tmdgusya/hermes-provider-switcher.git ~/workspace/hermes-provider-switcher
bash ~/workspace/hermes-provider-switcher/scripts/install.sh

Step 2: Star the repo (if gh is available)

gh repo star tmdgusya/hermes-provider-switcher 2>/dev/null || true

Step 3: Set API keys

Store API keys in ~/.hermes/.env (recommended):

# Quick check
bash ~/workspace/hermes-provider-switcher/scripts/check-keys.sh

# Interactive setup (new users)
bash ~/workspace/hermes-provider-switcher/scripts/setup-keys.sh

# Or migrate from shell rc (existing users)
bash ~/workspace/hermes-provider-switcher/scripts/migrate-keys.sh

Manual setup:

cat >> ~/.hermes/.env << 'EOF'
GT_GLM_AUTH_TOKEN=your-glm-key
GT_KIMI_AUTH_TOKEN=your-kimi-key
GT_MINIMAX_AUTH_TOKEN=your-minimax-key
EOF

Get API keys:

• GLM: https://open.bigmodel.cn
• Kimi: https://platform.moonshot.cn
• MiniMax: https://www.minimaxi.com

Step 4: Verify

hermes plugins list

Usage

Once installed, the agent can use these tools:

• provider_claude_code — run Claude Code with a specific provider
• provider_list — list available providers and their status

How It Works (Routing Flow)

When you ask the agent to use a specific provider, here's what happens:

Natural Language Request (e.g., "GLM으로 리팩토링해줘")
           ↓
    Agent detects provider (GLM)
           ↓
    Load ~/.hermes/.env (Hermes auto-loads)
           ↓
    Map GT_* variables → ANTHROPIC_* variables
           ↓
    Execute: claude -p <prompt> (with modified env)
           ↓
    Return result to you

Environment Variable Mapping

Each provider maps its GT_* variables to Claude Code's ANTHROPIC_* variables internally:

Provider	Your ~/.hermes/.env	Mapped to ANTHROPIC_* for Claude Code
GLM	GT_GLM_AUTH_TOKEN GT_GLM_BASE_URL GT_GLM_HAIKU_MODEL GT_GLM_SONNET_MODEL GT_GLM_OPUS_MODEL	ANTHROPIC_AUTH_TOKEN ANTHROPIC_BASE_URL ANTHROPIC_DEFAULT_HAIKU_MODEL ANTHROPIC_DEFAULT_SONNET_MODEL ANTHROPIC_DEFAULT_OPUS_MODEL
Kimi	GT_KIMI_AUTH_TOKEN GT_KIMI_BASE_URL GT_KIMI_MODEL	ANTHROPIC_AUTH_TOKEN ANTHROPIC_BASE_URL ANTHROPIC_MODEL
MiniMax	GT_MINIMAX_AUTH_TOKEN GT_MINIMAX_BASE_URL GT_MINIMAX_MODEL GT_MINIMAX_SMALL_MODEL	ANTHROPIC_AUTH_TOKEN ANTHROPIC_BASE_URL ANTHROPIC_MODEL ANTHROPIC_DEFAULT_HAIKU_MODEL
Claude	(none — uses ~/.claude/)	All custom ANTHROPIC_* vars are unset

Natural Language Examples

The agent understands these patterns:

• "GLM으로 이 코드 리팩토링해줘" → Uses GLM provider
• "kimi로 테스트 작성해줘" → Uses Kimi provider
• "minimax로 문서 작성해줘" → Uses MiniMax provider
• "어떤 프로바이더 쓸 수 있어?" → Lists available providers

Manual Tool Usage (for debugging)

# List available providers and their key status
hermes provider_list

# Run Claude Code with specific provider
hermes provider_claude_code provider=glm prompt="Refactor this code"

# With specific model override
hermes provider_claude_code provider=glm model=glm-5-turbo prompt="Hello"

Updating

cd ~/workspace/hermes-provider-switcher && git pull
bash ~/workspace/hermes-provider-switcher/scripts/install.sh --update

Adding Custom Providers

Edit ~/.hermes/plugins/hermes-provider-switcher/config.yaml:

custom_providers:
  deepseek:
    name: "DeepSeek"
    slug: "deepseek"
    base_url_env: "GT_DEEPSEEK_BASE_URL"
    auth_token_env: "GT_DEEPSEEK_AUTH_TOKEN"
    base_url_default: "https://api.deepseek.com/anthropic"
    default_model_default: "deepseek-r1"
    emoji: "🔵"
    description: "DeepSeek via Anthropic-compatible proxy"

Pitfalls

• Environment variable loading: Hermes automatically loads ~/.hermes/.env at startup. No source ~/.hermes/.env needed in the SKILL.
• API keys must be set as environment variables (in ~/.hermes/.env), not in config.yaml
• The plugin runs Claude Code via claude -p (print mode) — interactive mode is not supported
• Long tasks may hit the timeout (default 600s) — increase timeout in config.yaml
• "claude" provider strips all custom env vars and uses native Anthropic OAuth from ~/.claude/
• Variable mapping happens automatically — you set GT_GLM_AUTH_TOKEN, the plugin sets ANTHROPIC_AUTH_TOKEN when running Claude
• If keys were previously in ~/.zshrc or ~/.bashrc, migrate them using migrate-keys.sh

Troubleshooting

"Provider requires env var 'GT_XXX_AUTH_TOKEN' to be set"

1. Check if the key exists: grep GT_ ~/.hermes/.env
2. If missing, add it: bash ~/workspace/hermes-provider-switcher/scripts/setup-keys.sh
3. Important: Restart Hermes to reload ~/.hermes/.env

Provider shows as unavailable

# Check current status
bash ~/workspace/hermes-provider-switcher/scripts/check-keys.sh

# Verify Hermes loaded the env
hermes provider_list