Adapters¶

LLM provider adapters for unified API access.

Overview¶

Adapters provide a consistent interface to different LLM providers, handling the translation between RobotLab's message format and provider-specific APIs.

# Configure globally
RobotLab.configure do |config|
  config.default_model = "claude-sonnet-4"
  # Adapter is selected automatically based on model
end

# Or configure per-robot
robot = RobotLab.build do
  model "gpt-4o"  # Uses OpenAI adapter
end

Adapter Selection¶

Adapters are automatically selected based on model name:

Model Pattern	Adapter
`claude-`, `anthropic/`	Anthropic
`gpt-`, `o1-`, `openai/*`	OpenAI
`gemini-`, `google/`	Gemini

Common Interface¶

All adapters implement:

adapter.chat(
  messages: messages,
  model: model,
  tools: tools,
  system: system_prompt,
  streaming: callback
)
# => Response with content and usage

Available Adapters¶

Adapter	Description
Anthropic	Claude models via Anthropic API
OpenAI	GPT models via OpenAI API
Gemini	Gemini models via Google AI

Configuration¶

API Keys¶

Set via environment variables:

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

Custom Endpoints¶

RobotLab.configure do |config|
  config.adapter_options = {
    anthropic: { base_url: "https://custom.anthropic.endpoint" },
    openai: { base_url: "https://custom.openai.endpoint" }
  }
end

Creating Custom Adapters¶

Implement the adapter interface:

class MyAdapter
  def chat(messages:, model:, tools: [], system: nil, streaming: nil)
    # Translate messages to provider format
    # Make API call
    # Translate response back

    Response.new(
      content: content,
      tool_calls: tool_calls,
      usage: { input_tokens: x, output_tokens: y }
    )
  end
end

# Register the adapter
RobotLab.register_adapter(:my_provider, MyAdapter)