LLM Provider Configuration

Beamlens supports the vast majority of LLM providers. You can configure them in the client_registry section of your configuration.

{Beamlens,
  client_registry: %{
    primary: "Ollama",
    clients: [
      %{
        name: "Ollama",
        provider: "openai-generic",
        options: %{base_url: "http://localhost:11434/v1", model: "llama3.2"}
      }
    ]
  }}

Beamlens utilizes BAML to provide LLM Client support. See the ClientRegistry documentation if you have questions beyond what this documentation provides.

Default: Anthropic

By default Anthropic is configured with the claude-haiku-4-5-20251001 model. No configuration needed. Set your API key and you're ready:

export ANTHROPIC_API_KEY="sk-ant-..."

Anthropic

export ANTHROPIC_API_KEY="sk-ant-..."

client_registry: %{
  primary: "Anthropic",
  clients: [
    %{
      name: "Anthropic",
      provider: "anthropic",
      options: %{model: "claude-haiku-4-5-20251001"}
    }
  ]
}

Refer to the BAML Anthropic documentation for more details.

AWS Bedrock

AWS Bedrock uses standard AWS authentication methods.

client_registry: %{
  primary: "Bedrock",
  clients: [
    %{
      name: "Bedrock",
      provider: "aws-bedrock",
      options: %{
        model: "anthropic.claude-3-5-haiku-20241022-v1:0",
        region: "us-east-1"
      }
    }
  ]
}

Refer to the BAML AWS Bedrock documentation for more details.

Google AI Gemini

export GOOGLE_API_KEY="..."

client_registry: %{
  primary: "GoogleAI",
  clients: [
    %{
      name: "GoogleAI",
      provider: "google-ai",
      options: %{model: "gemini-2.0-flash"}
    }
  ]
}

Refer to the BAML Google AI Gemini documentation for more details.

Google Vertex AI

Uses Google Cloud Application Default Credentials.

client_registry: %{
  primary: "GoogleVertexAI",
  clients: [
    %{
      name: "GoogleVertexAI",
      provider: "vertex-ai",
      options: %{
        model: "gemini-2.5-pro",
        location: "us-central1",
        project_id: "my-project-id",
        query_params: %{
          key: "Your VERTEX_API_KEY" # Or set the API KEY
        }
      }
    }
  ]
}

Refer to the BAML Vertex AI documentation for more details.

Ollama (Local)

No API key. Your data stays on your machine.

# Install and run Ollama
ollama serve
ollama pull llama3.2

client_registry: %{
  primary: "Ollama",
  clients: [
    %{
      name: "Ollama",
      provider: "openai-generic",
      options: %{
        base_url: "http://localhost:11434/v1",
        model: "llama3.2"
      }
    }
  ]
}

Refer to the BAML Ollama documentation for more details.

OpenAI

export OPENAI_API_KEY="sk-..."

client_registry: %{
  primary: "OpenAI",
  clients: [
    %{
      name: "OpenAI",
      provider: "openai",
      options: %{model: "gpt-4o-mini"}
    }
  ]
}

Refer to the BAML OpenAI documentation for more details.

OpenAI Azure

export AZURE_OPENAI_API_KEY="..."

client_registry: %{
  primary: "AzureOpenAI",
  clients: [
    %{
      name: "AzureOpenAI",
      provider: "azure-openai",
      options: %{
        resource_name: "your-resource-name",
        deployment_id: "your-deployment-id",
        api_version: "2024-02-15-preview"
      }
    }
  ]
}

Refer to the BAML OpenAI Azure documentation for more details.

OpenAI Generic

client_registry: %{
  primary: "OpenAIGeneric",
  clients: [
    %{
      name: "OpenAIGeneric",
      provider: "openai-generic",
      options: %{
        base_url: "https://api.provider.com",
        model: "o<provider-specified-format>"
      }
    }
  ]
}

Refer to the BAML OpenAI Generic documentation for more details.

OpenRouter

export OPENROUTER_API_KEY="..."

client_registry: %{
  primary: "OpenRouter",
  clients: [
    %{
      name: "OpenRouter",
      provider: "openrouter",
      options: %{
        model: "openai/gpt-4o-mini"
      }
    }
  ]
}

Refer to the BAML OpenRouter documentation for more details.

Advanced Patterns

Retry Policy

Handle transient failures with exponential backoff. Beamlens includes a DefaultRetry policy (3 retries, exponential backoff starting at 200ms):

client_registry: %{
  primary: "Anthropic",
  clients: [
    %{
      name: "Anthropic",
      provider: "anthropic",
      options: %{model: "claude-haiku-4-5-20251001"},
      retry_policy: "DefaultRetry"
    }
  ]
}

The DefaultRetry policy uses:

• 3 retries after initial failure
• Exponential backoff starting at 200ms
• 1.5x multiplier per retry
• Maximum delay of 10 seconds

Fallback Chains

Switch providers when your primary fails:

client_registry: %{
  primary: "PrimaryWithFallback",
  clients: [
    %{
      name: "Primary",
      provider: "anthropic",
      options: %{model: "claude-haiku-4-5-20251001"}
    },
    %{
      name: "Backup",
      provider: "openai",
      options: %{model: "gpt-4o-mini"}
    },
    %{
      name: "PrimaryWithFallback",
      provider: "fallback",
      options: %{strategy: ["Primary", "Backup"]}
    }
  ]
}

Round-Robin

Distribute load across multiple providers or API keys:

client_registry: %{
  primary: "RoundRobinClient",
  clients: [
    %{
      name: "Client1",
      provider: "anthropic",
      options: %{model: "claude-haiku-4-5-20251001"}
    },
    %{
      name: "Client2",
      provider: "anthropic",
      options: %{model: "claude-haiku-4-5-20251001"}
    },
    %{
      name: "RoundRobinClient",
      provider: "round-robin",
      options: %{strategy: ["Client1", "Client2"]}
    }
  ]
}

Each request cycles through the clients in order. Useful for spreading rate limits across multiple API keys.

More Providers

Beamlens uses BAML for LLM integration. Any provider BAML supports works with Beamlens:

• Together AI
• Groq
• Vercel AI Gateway
• And more

See the BAML provider documentation for the full list and configuration options.