Source code for lalandre_rag.summaries.agent

"""PydanticAI agent for structured canonical summary generation."""

from __future__ import annotations

from collections.abc import Callable
from typing import Any

from lalandre_core.config import get_config
from lalandre_core.llm.structured import run_structured_agent
from pydantic import BaseModel, Field, field_validator
from pydantic_ai import Agent, ModelRetry




class CanonicalSummaryOutput(BaseModel):
    """Structured summary output validated by PydanticAI."""

    model_config = {"extra": "ignore"}

    summary: str = Field(min_length=10)
    output_validation_retries: int = Field(default=0, ge=0)



    @field_validator("summary", mode="before")
    @classmethod
    def clean_summary(cls, v: Any) -> str:
        """Normalize and validate the structured summary text."""
        if isinstance(v, str) and v.strip():
            return v.strip()
        raise ValueError("Summary cannot be empty")




_SUMMARY_AGENT: Agent[Any, CanonicalSummaryOutput] = Agent(
    output_type=CanonicalSummaryOutput,
    instructions=(
        "Tu résumes un acte juridique pour un panneau de bibliothèque documentaire.\n"
        "4 à 6 phrases maximum. Ton factuel, sans citations, sans markdown, sans liste.\n"
        "Décris l'objet du texte, sa portée et sa structure si visible.\n"
        "N'invente aucun élément absent du contenu fourni."
    ),
    output_retries=2,
    defer_model_check=True,
)


@_SUMMARY_AGENT.output_validator
def _validate_summary(result: CanonicalSummaryOutput) -> CanonicalSummaryOutput:
    min_chars = get_config().search.summary_min_chars
    if len(result.summary) < min_chars:
        raise ModelRetry(f"Le résumé est trop court (min {min_chars} caractères).")
    return result




def run_summary_agent(
    *,
    prompt: str,
    generate_text: Callable[[str], str],
    model_name: str,
) -> tuple[CanonicalSummaryOutput, int]:
    """Run the canonical summary agent and return validated output + retry count."""
    return run_structured_agent(
        agent=_SUMMARY_AGENT,
        prompt=prompt,
        llm_or_generate=generate_text,
        model_name=model_name,
    )