core.types

Tool types

Tool-related models:

• ToolSpec: name + optional description + JSON schema parameters
• ToolCall: tool call request (id, name, arguments dict)

Prompt types

Prompt-related models:

• PromptPayload: prompt text or messages, plus optional prompt_id, prompt_version, prompt_hash

LLM types

LLM-related models:

• LLMConfig: model + tools + generation settings
• LLMRequest: prompt + config
• LLMResponse: text + tool calls + raw + usage
• LLMStreamChunk: streaming delta type

Provider protocols

Protocols (interfaces):

• LLMProvider: complete(request) -> LLMResponse
• StreamingLLMProvider: adds stream(request) -> Iterator[LLMStreamChunk]

Next steps

LLMs and providers

How these types are used in llm_step and providers.

OpenAI provider reference

A concrete mapping of these types to a provider implementation.