Using Anthropic with Lumen AI#

Prerequisites#

• Lumen AI installed in your Python environment.

• An Anthropic API Key. You can obtain this from the Anthropic Console.

Using Environment Variables#

By default Lumen will pick the first LLM provider it can find an API key for. By setting it you can give it access to that provider.

1. Set the Environment Variable

Set the OPENAI_API_KEY environment variable in your system. This allows Lumen AI to automatically detect and use Anthropic as the provider.

export ANTHROPIC_API_KEY='your-anthropic-api-key'

set ANTHROPIC_API_KEY=your-anthropic-api-key

2. Simply run your Lumen AI application, and it will automatically use Anthropic as the LLM provider. If you have environment variables for multiple providers override it.

lumen-ai serve <your-data-file-or-url> [--provider anthropic]

Using CLI Arguments#

Alternatively you can also provide the API key as a CLI argument:

lumen-ai serve <your-data-file-or-url> --provider anthropic --api-key <your-anthropic-api-key>

Python#

In Python, simply import the LLM wrapper lumen.ai.llm.AnthropicAI and pass it to the lumen.ai.ui.ExplorerUI:

import lumen.ai as lmai

anthropic_llm = lmai.llm.AnthropicAI(api_key='your-anthropic-api-key')

ui = lmai.ui.ExplorerUI('<your-data-file-or-url>', llm=anthropic_llm)
ui.servable()

Configuring the Model#

If you want to override the default and/or reasoning models, which use Claude 3 Haiku and Claude 3.5 Sonnet respectively, a model configuration can be provided via the model_kwargs parameter. Lumen will use different models for different scenarios, currently a default and a reasoning model can be provided. If no reasoning model is provided it will always use the default model.

You can find the latest list of Anthropic models in their documentation. To swap Opus out for Haiku as the default model simply override the default model_kwargs:

import lumen.ai as lmai

config = {
    "default": "claude-3-opus-latest",
    "reasoning": "claude-3-5-sonnet-latest"
}

llm = lmai.llm.AnthropicAI(model_kwargs=config)

lmai.ui.ExplorerUI('<your-data-file-or-url>', llm=llm).servable()