app/ask.py

"""
CLI client — retrieve + answer in one command, no server.

Useful for:
  - quick smoke-testing the bot before standing up the API
  - testing one question without keeping a server alive
  - sanity-checking that retrieval is finding the right passages

Usage:
    python -m app.ask "Can a non-Irish company apply for funding?"
    python -m app.ask --no-llm "..."         # just show top-k passages
    python -m app.ask --top-k 8 "..."
"""

from __future__ import annotations

import argparse
import asyncio
import sys

from app.config import SETTINGS
from app.prompt import IDK_MESSAGE, SYSTEM_PROMPT, format_user_message
from app.providers.factory import build_llm
from app.retrieve import retrieve


async def _ask(question: str, top_k: int, no_llm: bool) -> int:
    print(f"\nQ: {question}\n", flush=True)

    result = retrieve(question, top_k=top_k)

    print(f"-- top-{top_k} passages (best_score={result.best_score:.3f}, "
          f"threshold={SETTINGS.relevance_threshold}, "
          f"gate_{'PASS' if result.gate_passed else 'FAIL'}) --", flush=True)
    for i, p in enumerate(result.passages, 1):
        page = f" p.{p.page}" if p.page else ""
        head = (p.heading_path or "").replace("\n", " ")[:70]
        print(f"  [{i}] {p.score:.3f} | {p.category:8s} | {head} | {p.source_url}{page}",
              flush=True)

    if no_llm:
        return 0 if result.gate_passed else 2

    if not result.gate_passed:
        print(f"\nA: {IDK_MESSAGE}\n", flush=True)
        return 2

    try:
        llm = build_llm(SETTINGS)
    except Exception as e:
        print(f"\n[error] could not build LLM provider: {e}", file=sys.stderr)
        return 3

    print(f"\nA ({llm.name}): ", end="", flush=True)
    messages = [
        {"role": "system", "content": SYSTEM_PROMPT},
        {"role": "user", "content": format_user_message(question, result.passages)},
    ]
    try:
        async for tok in llm.stream(messages):
            print(tok, end="", flush=True)
    except Exception as e:
        print(f"\n[error] LLM stream failed: {e}", file=sys.stderr)
        return 4
    print("\n", flush=True)

    # Show deduplicated source URLs
    seen, sources = set(), []
    for p in result.passages:
        if p.source_url in seen:
            continue
        seen.add(p.source_url)
        sources.append(p)
    print("Sources:", flush=True)
    for s in sources:
        page = f" (page {s.page})" if s.page else ""
        print(f"  - {s.title}{page}\n    {s.source_url}", flush=True)
    return 0


def main() -> int:
    parser = argparse.ArgumentParser()
    parser.add_argument("question", help="Your question (in quotes)")
    parser.add_argument("--top-k", type=int, default=SETTINGS.top_k)
    parser.add_argument("--no-llm", action="store_true",
                        help="Skip LLM call; only print retrieved passages.")
    args = parser.parse_args()
    return asyncio.run(_ask(args.question, args.top_k, args.no_llm))


if __name__ == "__main__":
    raise SystemExit(main())