AI/ontology_platform/ont_platform/config.py

"""Platform-level configuration.

Thin layer over OntoCast's ``Config`` that adds:

1. A single explicit ``PHASE`` knob. Phase 0 forces filesystem-only storage
   and ignores any Neo4j/Fuseki environment variables that may be present
   (they belong to Phase 4 — see ``docs/통합설계서.md`` §5).
2. A platform-side ``working_directory`` default so the user does not have
   to remember the legacy ``ONTOCAST_*`` env-var prefix.
3. Hook points for future phases (LLM budgets, robots policy, etc.).

The OntoCast vendored copy itself is not modified by this module — we only
build a ``Config`` instance and pass it to ``ToolBox``.
"""

from __future__ import annotations

import logging
import os
import sys
from enum import IntEnum
from pathlib import Path
from typing import Literal

from pydantic import Field, model_validator
from pydantic_settings import BaseSettings, SettingsConfigDict

# Make the vendored OntoCast package importable without users having to
# install it into the environment. This mirrors what tests/unit/test_*.py
# already do; centralizing it here keeps any module that imports our
# Settings consistent.
_REPO_ROOT = Path(__file__).resolve().parents[1]
_VENDORED_ONTOCAST = _REPO_ROOT / "vendored" / "ontocast"
if str(_VENDORED_ONTOCAST) not in sys.path:
    sys.path.insert(0, str(_VENDORED_ONTOCAST))

# Late import — OntoCast must be on sys.path first.
from ontocast.config import (  # noqa: E402  (must follow sys.path tweak)
    Config as OntoCastConfig,
)
from ontocast.config import (  # noqa: E402
    FusekiConfig,
    LLMConfig,
    Neo4jConfig,
    PathConfig,
    ToolConfig,
)

logger = logging.getLogger(__name__)


class Phase(IntEnum):
    """Integration phase from the master design (docs/통합설계서.md §5).

    Used as a gate: features whose phase exceeds ``Settings.phase`` are
    disabled even if their environment variables are populated.
    """

    BASE = 0  # OntoCast only, filesystem storage
    TRAFILATURA = 1
    CANDIDATE_REVIEW = 2
    CRAWL4AI = 3
    GUARDRAILS = 4
    NEO4J_GRAPHRAG = 5
    MULTI_AGENT = 6


StorageBackend = Literal["filesystem", "fuseki", "neo4j"]


class PlatformSettings(BaseSettings):
    """Top-level platform settings, loaded from ``.env`` and the environment.

    Only fields the platform itself needs are declared here. OntoCast-specific
    settings (LLM, chunk, paths, ...) are loaded by OntoCast's own ``Config``
    via its own ``env_prefix`` conventions and merged in ``build_ontocast_config``.
    """

    model_config = SettingsConfigDict(
        env_file=".env",
        env_file_encoding="utf-8",
        extra="ignore",
        case_sensitive=False,
    )

    # ─── Phase gating ────────────────────────────────────────────────
    phase: Phase = Field(
        default=Phase.BASE,
        description="Current integration phase (docs/통합설계서.md §5).",
    )
    storage_backend: StorageBackend = Field(
        default="filesystem",
        description=(
            "Which storage backend to use. Phase 0 forces 'filesystem' "
            "regardless of any Neo4j/Fuseki credentials in the environment."
        ),
    )
    database_url: str = Field(
        default="sqlite:///./data/ontology_platform.db",
        description="SQLAlchemy database URL for Phase 2 candidate/review storage.",
    )

    # ─── Paths (mirror ONTOCAST_* but with platform defaults) ────────
    working_directory: Path = Field(
        default=Path("./data/working"),
        description="Working directory for OntoCast filesystem store.",
        validation_alias="ONTOCAST_WORKING_DIRECTORY",
    )
    ontology_directory: Path | None = Field(
        default=None,
        validation_alias="ONTOCAST_ONTOLOGY_DIRECTORY",
    )

    # ─── Server ──────────────────────────────────────────────────────
    host: str = Field(default="0.0.0.0")
    port: int = Field(default=8000)
    log_level: str = Field(default="info")

    # ─── Operational policy (used from Phase 2+) ─────────────────────
    robots_policy: Literal["strict", "respect", "ignore"] = Field(
        default="respect",
        description="robots.txt 준수 정책. Phase 2 Crawl4AI 통합에서 사용.",
    )
    crawler_default_profile: Literal[
        "fast_static",
        "dynamic_page",
        "full_capture",
        "structured_extract",
        "deep_discovery",
    ] = Field(default="fast_static")
    crawler_cache_policy: Literal["enabled", "disabled", "bypass"] = Field(default="enabled")
    crawler_max_pages: int = Field(default=50, ge=1, le=50)
    crawler_max_depth: int = Field(default=1, ge=0, le=5)
    text2cypher_result_limit: int = Field(default=100, ge=1, le=1000)
    graph_search_result_limit: int = Field(default=20, ge=1, le=100)
    daily_llm_call_limit: int = Field(default=10_000)
    daily_llm_token_limit: int = Field(default=10_000_000)

    # ─── Phase 0 enforcement ─────────────────────────────────────────
    @model_validator(mode="after")
    def _enforce_phase_storage_consistency(self) -> PlatformSettings:
        """Phase 0 forces filesystem; later phases may opt into other backends.

        Anything other than 'filesystem' before Phase 4 is treated as a
        configuration error — failing fast here is far easier to debug than
        a half-wired Neo4j connection deep inside ToolBox.
        """
        if self.phase < Phase.NEO4J_GRAPHRAG and self.storage_backend != "filesystem":
            raise ValueError(
                f"storage_backend={self.storage_backend!r} requires Phase 4+/Phase 5+ "
                f"(Phase 5 in the current roadmap), "
                f"but PHASE={int(self.phase)}. See docs/통합설계서.md §5."
            )
        # Ensure working directory exists for filesystem mode.
        if self.storage_backend == "filesystem":
            self.working_directory.mkdir(parents=True, exist_ok=True)
        return self


def _build_llm_config_lenient() -> LLMConfig:
    """env vars로부터 LLMConfig 생성. OntoCast의 OpenAIModel enum validation을 우회한다.

    Why: LM Studio / vLLM / llama.cpp 등 OpenAI-호환 로컬 서버는 임의의 model
    identifier를 쓰며(예: `deepseek-r1-distill-qwen-7b`), 이는 OntoCast의 정해진
    enum(`gpt-4o`, `gpt-4o-mini`, ...)에 들어가지 않는다. ChatOpenAI는 model을
    문자열로 받으므로 enum 강제만 풀면 OntoCast 다른 코드 경로는 그대로 동작한다.

    `LLMConfig.model_construct`는 Pydantic V2의 validation 우회 생성자다.
    """
    provider_raw = (os.getenv("LLM_PROVIDER") or "openai").lower()
    model_name = os.getenv("LLM_MODEL_NAME") or "gpt-4o-mini"
    temperature_raw = os.getenv("LLM_TEMPERATURE") or "0.0"
    base_url = os.getenv("LLM_BASE_URL") or None
    api_key = os.getenv("LLM_API_KEY") or None

    return LLMConfig.model_construct(
        provider=provider_raw,
        model_name=model_name,
        temperature=float(temperature_raw),
        base_url=base_url,
        api_key=api_key,
    )


def _empty_neo4j_config() -> Neo4jConfig:
    """A Neo4jConfig with no URI/auth so ToolBox skips Neo4j initialization.

    ToolBox enables Neo4j only when both ``uri`` and ``auth`` are set
    (vendored/ontocast/ontocast/toolbox.py around line 112). Returning an
    instance with both ``None`` is the canonical way to disable the backend
    even if NEO4J_* env vars happen to be populated in the shell.
    """
    cfg = Neo4jConfig()
    cfg.uri = None  # type: ignore[assignment]
    cfg.auth = None  # type: ignore[assignment]
    return cfg


def _empty_fuseki_config() -> FusekiConfig:
    """Same idea as ``_empty_neo4j_config`` for Fuseki."""
    cfg = FusekiConfig()
    cfg.uri = None  # type: ignore[assignment]
    cfg.auth = None  # type: ignore[assignment]
    return cfg


def build_ontocast_config(settings: PlatformSettings) -> OntoCastConfig:
    """Build an OntoCast ``Config`` from our ``PlatformSettings``.

    Phase 0 behavior:
    - LLM/Chunk/Aggregation/WebSearch sections come from their own env vars
      via OntoCast's ``BaseSettings`` defaults. We do NOT touch them.
    - Paths are overridden so the OntoCast working directory always matches
      ``PlatformSettings.working_directory``.
    - Neo4j and Fuseki are forcibly disabled regardless of NEO4J_*/FUSEKI_*
      env vars in the shell. They will be wired in Phase 4.
    """
    # OntoCast의 OpenAIModel enum은 클라우드 모델만 허용한다. LM Studio 등
    # 임의의 모델명을 쓰는 로컬 서버는 ToolConfig() 생성 단계에서 검증이 실패
    # 하므로, ToolConfig를 만들 동안만 LLM_MODEL_NAME을 비우고 lenient 빌더로
    # 교체한다. CHUNK_*/AGG_* 등 다른 섹션 env는 그대로 흘러가도록 유지한다.
    saved_model = os.environ.pop("LLM_MODEL_NAME", None)
    try:
        tool_cfg = ToolConfig()
    finally:
        if saved_model is not None:
            os.environ["LLM_MODEL_NAME"] = saved_model

    tool_cfg.llm_config = _build_llm_config_lenient()

    # Override paths from the platform settings.
    tool_cfg.path_config = PathConfig(
        working_directory=settings.working_directory,
        ontology_directory=settings.ontology_directory,
    )

    # Phase 0: forcibly disable non-filesystem backends.
    if settings.storage_backend == "filesystem":
        tool_cfg.neo4j = _empty_neo4j_config()
        tool_cfg.fuseki = _empty_fuseki_config()
    # (No 'else' here yet — Phase 4 will add the fuseki/neo4j branches.)

    cfg = OntoCastConfig(tool_config=tool_cfg)

    # Validate LLM config eagerly so misconfiguration surfaces at startup,
    # not on the first /process call.
    try:
        cfg.validate_llm_config()
    except ValueError as exc:
        # Re-raise with a hint about where to set the variable. OntoCast's
        # error already mentions the env var name; we add the file location.
        raise ValueError(f"{exc} (set it in `.env` or your shell environment)") from exc

    return cfg


def load_settings() -> PlatformSettings:
    """Load `PlatformSettings` from the environment.

    Separated from module-level instantiation so tests can swap env vars
    via ``monkeypatch`` without import-time side effects.
    """
    return PlatformSettings()  # type: ignore[call-arg]


# Provider-grade LLM/embedding config helpers can be added in later phases.
# For Phase 0, OntoCast's own ``LLMConfig`` is sufficient.
__all__ = [
    "LLMConfig",
    "Phase",
    "PlatformSettings",
    "StorageBackend",
    "build_ontocast_config",
    "load_settings",
]