Reliability is a configuration, not code

Models fail. Rate limits hit. BAML handles this declaratively—no try/catch spaghetti.

Define retry policies and fallback chains. Swap models without touching your functions.

main.baml
retry_policy Resilient {
  max_retries 3
  strategy {
    type exponential_backoff
    initial_delay_ms 500
    max_delay_ms 10000
  }
}

client<llm> ReliableGPT {
  provider openai
  options {
    model "gpt-4o-mini"
    api_key env.OPENAI_API_KEY
  }
  retry_policy Resilient
}

client<llm> FallbackChain {
  provider "baml-fallback"
  options {
    strategy [
      ReliableGPT,
      { provider anthropic, options { model "claude-3-haiku-20240307" } }
    ]
  }
}
Click Run to see the rendered prompt
← StreamingIntegration →