src/tools/wikipedia_tool.py

import json
import httpx
from langchain_core.tools import tool
from langchain_community.utilities import WikipediaAPIWrapper

_REST_BASE = "https://en.wikipedia.org/api/rest_v1/page/summary"
_HEADERS = {"User-Agent": "logic-assistant/1.0 (educational tool)"}
_TIMEOUT = 8

_wrapper = WikipediaAPIWrapper(top_k_results=1, doc_content_chars_max=1500, lang="en")


def _rest_summary(concept: str) -> str | None:
    """
    Try the Wikipedia REST summary endpoint.
    Returns extract text, or None if the page isn't found.
    Raises on network/server errors so the caller can fall back.
    """
    slug = concept.strip().replace(" ", "_")
    url = f"{_REST_BASE}/{httpx.URL(slug)}"
    resp = httpx.get(url, headers=_HEADERS, timeout=_TIMEOUT, follow_redirects=True)
    if resp.status_code == 404:
        return None
    resp.raise_for_status()
    return resp.json().get("extract") or None


@tool
def wikipedia(query: str) -> str:
    """Fetch a public-domain definition of a logic or philosophy concept from Wikipedia.
    Input must be the English concept name (e.g. 'hasty generalization', 'ad hominem').
    Call this for every key concept found by search_graph so we have an external definition."""
    # 1. Try REST summary API — fast, reliable, no JSONDecodeError risk
    try:
        text = _rest_summary(query)
        if text:
            return text
    except (httpx.HTTPStatusError, httpx.RequestError):
        pass  # fall through to wrapper

    # 2. Fallback: WikipediaAPIWrapper (search-based, handles fuzzy titles)
    try:
        return _wrapper.run(query)
    except json.JSONDecodeError:
        raise RuntimeError(
            f"Wikipedia API returned an empty response for '{query}' "
            "(transient network issue). Use tavily_search as fallback."
        )