Providers

The engine supports three provider adapters through a common interface.

Supported Providers

• anthropic
• openai
• gemini

There are no default models. A model must be specified explicitly via --model (CLI) or model (API request); omitting it raises an error.

Env Vars

• ANTHROPIC_API_KEY
• OPENAI_API_KEY
• GOOGLE_API_KEY

Unified Interface

All providers implement:

• complete(messages, tools, temperature, max_tokens, json_mode)
• complete_json(...)

Unified response fields:

• content
• tool_calls
• input_tokens
• output_tokens
• model
• stop_reason

Provider-Specific Notes

Anthropic

• Uses anthropic.AsyncAnthropic.
• JSON mode is steered via a system-prompt instruction (the model is asked to return only a JSON object), and markdown code fences are stripped from the response. Assistant-message prefill is not used — current Claude models reject it (the conversation must end with a user message).

OpenAI

• Uses openai.AsyncOpenAI chat completions.
• JSON mode uses response_format={"type": "json_object"}.

Gemini

• Uses google.genai client.
• JSON mode sets response_mime_type="application/json".

Unknown Provider Handling

get_provider(...) raises ValueError for unsupported provider names, and also when no model is supplied (there is no default model).