`lasagnastack.llm.base`¶

lasagnastack.llm.base ¶

LLMClient ¶

LLMClient(
    reasoning_max_tokens: int = 4000,
    reasoning_effort: str | None = None,
    total_max_tokens: int | None = None,
)

Bases: ABC

Abstract base class for all LLM provider clients.

Subclasses must implement :meth:generate and :meth:generate_with_video. The constructor stores three shared reasoning-control parameters that each concrete client interprets according to its provider's API:

reasoning_max_tokens — a hard token budget for internal reasoning/thinking. Gemini maps this to thinking_budget; OpenRouter maps it to reasoning.max_tokens (when no effort level is set).
reasoning_effort — a qualitative effort level ("none" / "minimal" / "low" / "medium" / "high" / "xhigh"). OpenRouter maps this to reasoning.effort and it takes priority over reasoning_max_tokens when both are supplied. Not used by Gemini.
total_max_tokens — ceiling on all output tokens combined (reasoning + visible response). Maps directly to the max_tokens API parameter. Required for Anthropic models via OpenRouter, where this total must be strictly greater than the reasoning budget. Ignored by Gemini.

Store shared reasoning-control parameters.

Parameters:

Name	Type	Description	Default
`reasoning_max_tokens`	`int`	Maximum tokens the model may spend on internal reasoning per call. Set to `0` to disable reasoning entirely (when no effort level overrides it).	`4000`
`reasoning_effort`	`str \| None`	Qualitative effort level controlling reasoning depth. When set, takes priority over `reasoning_max_tokens` for providers that support it (e.g. OpenRouter). `None` defers to `reasoning_max_tokens`.	`None`
`total_max_tokens`	`int \| None`	Hard ceiling on all output tokens combined (reasoning + visible response). Maps to the `max_tokens` API parameter. Required for Anthropic models (via OpenRouter) to ensure the total exceeds the reasoning budget. `None` lets the provider use its default.	`None`

Source code in src/lasagnastack/llm/base.py

def __init__(
    self,
    reasoning_max_tokens: int = 4000,
    reasoning_effort: str | None = None,
    total_max_tokens: int | None = None,
) -> None:
    """Store shared reasoning-control parameters.

    Args:
        reasoning_max_tokens: Maximum tokens the model may spend on
            internal reasoning per call. Set to ``0`` to disable
            reasoning entirely (when no effort level overrides it).
        reasoning_effort: Qualitative effort level controlling reasoning
            depth. When set, takes priority over ``reasoning_max_tokens``
            for providers that support it (e.g. OpenRouter). ``None``
            defers to ``reasoning_max_tokens``.
        total_max_tokens: Hard ceiling on all output tokens combined
            (reasoning + visible response). Maps to the ``max_tokens``
            API parameter. Required for Anthropic models (via OpenRouter)
            to ensure the total exceeds the reasoning budget. ``None``
            lets the provider use its default.
    """
    self._reasoning_max_tokens = reasoning_max_tokens
    self._reasoning_effort = reasoning_effort
    self._total_max_tokens = total_max_tokens

generate `abstractmethod` ¶

generate(
    prompt: str,
    response_schema: type[BaseModel],
    *,
    temperature: float = 0.4,
) -> BaseModel

Send a text-only prompt; return a validated response_schema instance.

Source code in src/lasagnastack/llm/base.py

@abstractmethod
def generate(
    self,
    prompt: str,
    response_schema: type[BaseModel],
    *,
    temperature: float = 0.4,
) -> BaseModel:
    """Send a text-only prompt; return a validated response_schema instance."""
    ...

generate_with_video `abstractmethod` ¶

generate_with_video(
    video_path: Path,
    prompt: str,
    response_schema: type[BaseModel],
    *,
    temperature: float = 0.4,
) -> BaseModel

Upload video_path to the Files API, generate, return validated response.

Source code in src/lasagnastack/llm/base.py

@abstractmethod
def generate_with_video(
    self,
    video_path: Path,
    prompt: str,
    response_schema: type[BaseModel],
    *,
    temperature: float = 0.4,
) -> BaseModel:
    """Upload video_path to the Files API, generate, return validated response."""
    ...

lasagnastack.llm.base¶