Model Failover Skill

Automatically switch between LLM providers when one fails. Supports configurable fallback chains, rate limiting, and health monitoring. Inspired by OpenClaw's model failover system.

Setup

Configure your provider chain in environment variables:

# 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

Environment Variable	Description	Default
LLM_PROVIDER_CHAIN	Comma-separated provider:model pairs	anthropic:claude-3-5-sonnet-20241022
ANTHROPIC_API_KEY	Anthropic API key	-
OPENAI_API_KEY	OpenAI API key	-
GOOGLE_API_KEY	Google API key	-
CUSTOM_API_KEY	Custom provider API key	-
MAX_RETRIES	Max retries per provider	2
RETRY_DELAY_MS	Delay between retries	1000

Provider Format

provider:model

Supported providers:

• 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