langgoap.GoalInterpreter

langgoap.GoalInterpreter#

class GoalInterpreter(llm, actions, system_prompt=None, structured_output_kwargs=None)[source]#

Converts natural language requests into GoalSpec objects via LLM.

Uses BaseChatModel.with_structured_output() to extract structured goals from free-text requests. Provider-agnostic: any chat model that supports structured output works.

Parameters:

  • llm (BaseChatModel) – A LangChain chat model (e.g. ChatOpenAI, ChatAnthropic).

  • actions (list[ActionSpec]) – Available GOAP actions (used to build the prompt catalog).

  • system_prompt (str | None) – Custom system prompt template. Must contain {action_catalog} and {world_state} placeholders. Defaults to DEFAULT_SYSTEM_PROMPT.

  • structured_output_kwargs (dict[str, Any] | None) – Optional keyword arguments forwarded to llm.with_structured_output(). Use this for provider-specific options, e.g. {"method": "function_calling"} for OpenAI when conditions must remain an open dict (OpenAI strict JSON-schema mode requires additionalProperties: false on all objects, which is incompatible with a free-form conditions mapping). When None (the default), auto-detects OpenAI models and applies {"method": "function_calling"} automatically.

__init__(llm, actions, system_prompt=None, structured_output_kwargs=None)[source]#
Parameters:

  • llm (BaseChatModel)

  • actions (list[ActionSpec])

  • system_prompt (str | None)

  • structured_output_kwargs (dict[str, Any] | None)

Return type:

None

Methods

__init__(llm, actions[, system_prompt, ...])

ainterpret(request[, world_state])

Async variant of interpret().

ainterpret_raw(request[, world_state])

Async variant of interpret_raw().

build_messages(request, world_state)

interpret(request[, world_state])

Interpret a natural language request and return a GoalSpec.

interpret_raw(request[, world_state])

Interpret a request and return the raw InterpretedGoal.