Round Robin

Distribute requests evenly across models:

workers:
  - name: round_robin
    type: load_balance
    load_balance:
      routing:
        - from: model_a
        - from: model_b
        - from: model_c

Fallback (Priority Order)

Try models in order, falling back on failure:

workers:
  - name: fallback
    type: load_balance
    load_balance:
      routing:
        - from: primary_model
          order: 1
        - from: backup_model
          order: 2
        - from: emergency_model
          order: 3

Weighted Distribution

Route by percentage weight:

workers:
  - name: weighted
    type: load_balance
    load_balance:
      routing:
        - from: fast_model
          weight: 8     # 80% of traffic
        - from: slow_model
          weight: 2     # 20% of traffic

Using Workers

Workers are invoked using the same API as models:

curl http://localhost:8090/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "fallback",
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Full Example

models:
  - from: openai:gpt-4o
    name: gpt4
    params:
      openai_api_key: ${ secrets:OPENAI_API_KEY }
  - from: anthropic:claude-sonnet-4-5
    name: claude
    params:
      anthropic_api_key: ${ secrets:ANTHROPIC_API_KEY }

workers:
  - name: smart_router
    type: load_balance
    description: Try GPT-4 first, fall back to Claude
    load_balance:
      routing:
        - from: gpt4
          order: 1
        - from: claude
          order: 2

Round Robin

Fallback (Priority Order)

Weighted Distribution

Using Workers

Full Example

Documentation

spice-workers

Spice Workers

Basic Configuration

Load Balancing Strategies