Model Failover Skill

Setup

# Comma-separated list of providers (in fallback order)
export LLM_PROVIDER_CHAIN="anthropic:claude-3-5-sonnet-20241022,openai:gpt-4o-mini,google:gemini-1.5-flash"

# API keys for each provider
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."

Usage

Chat with automatic failover

{baseDir}/model-failover.js chat "Your message here"

Add a new provider to the chain

{baseDir}/model-failover.js add-provider anthropic claude-3-5-sonnet-20241022

Remove a provider from the chain

{baseDir}/model-failover.js remove-provider openai

List providers in chain

{baseDir}/model-failover.js list

Check provider health

{baseDir}/model-failover.js health

Reset failure counts

{baseDir}/model-failover.js reset

Configuration

LLM_PROVIDER_CHAIN

provider:model

anthropic:claude-3-5-sonnet-20241022

ANTHROPIC_API_KEY

OPENAI_API_KEY

GOOGLE_API_KEY

CUSTOM_API_KEY

MAX_RETRIES

RETRY_DELAY_MS

Provider Format

provider:model

• anthropic

  - Anthropic Claude models

• openai

  - OpenAI GPT models

• google

  - Google Gemini models

• custom

  - Custom OpenAI-compatible endpoint (set

  OPENAI_BASE_URL

  )

How It Works

1. Try the first provider in the chain
2. If it fails (rate limit, error, timeout), wait and retry
3. If retries exhausted, move to next provider
4. Continue until success or all providers exhausted
5. Track failures per provider for health monitoring