app.py

import json
import streamlit as st

from corpus_loader import build_corpus
from dci_engine import search_corpus
from llm_planner import plan_search, synthesize_answer

st.set_page_config(page_title="DCI Agent", page_icon="🔍", layout="wide")
st.title("🔍 DCI Agent")
st.caption("Przeszukuj dokumenty bez embeddingów i baz wektorowych.")

# ── Sidebar ──────────────────────────────────────────────────────────────────
with st.sidebar:
    st.header("Konfiguracja")
    api_key = st.text_input("OpenAI API Key", type="password", placeholder="sk-...")

    col_btn, col_status = st.columns([1, 2])
    with col_btn:
        fetch_clicked = st.button("Pobierz modele", disabled=not api_key)
    if fetch_clicked:
        try:
            from openai import OpenAI as _OAI
            _models = _OAI(api_key=api_key).models.list()
            gpt_ids = sorted(
                [m.id for m in _models if "gpt" in m.id and "realtime" not in m.id and "audio" not in m.id],
                reverse=True,
            )
            st.session_state["available_models"] = gpt_ids
            with col_status:
                st.success(f"{len(gpt_ids)} modeli")
        except Exception as e:
            with col_status:
                st.error(str(e))

    _default_models = ["gpt-4o-mini", "gpt-4o"]
    _model_list = st.session_state.get("available_models", _default_models)
    model = st.selectbox("Model", _model_list)

    st.header("Korpus dokumentów")
    uploaded = st.file_uploader(
        "Wgraj pliki (.txt / .md / .pdf)",
        type=["txt", "md", "pdf"],
        accept_multiple_files=True,
    )

    if st.button("Załaduj korpus", disabled=not uploaded):
        import tempfile, os
        paths = []
        for f in uploaded:
            suffix = os.path.splitext(f.name)[1]
            with tempfile.NamedTemporaryFile(delete=False, suffix=suffix) as tmp:
                tmp.write(f.read())
                paths.append((tmp.name, f.name))

        corpus = {}
        for tmp_path, orig_name in paths:
            from corpus_loader import load_file
            try:
                text = load_file(tmp_path)
                corpus[orig_name] = text.splitlines()
            except Exception as e:
                st.warning(f"Nie można wczytać {orig_name}: {e}")
            finally:
                os.unlink(tmp_path)

        st.session_state["corpus"] = corpus
        total = sum(len(v) for v in corpus.values())
        st.success(f"Załadowano {len(corpus)} plik(ów), {total} linii.")

# ── Main area ─────────────────────────────────────────────────────────────────
corpus = st.session_state.get("corpus")
if corpus:
    st.info(f"Korpus: {len(corpus)} plik(ów) — " + ", ".join(corpus.keys()))

question = st.text_area("Pytanie", placeholder="Co chcesz wiedzieć z dokumentów?", height=80)

if st.button("🔎 Szukaj", type="primary"):
    if not api_key:
        st.error("Podaj klucz OpenAI API w panelu bocznym.")
    elif not corpus:
        st.error("Najpierw załaduj korpus w panelu bocznym.")
    elif not question.strip():
        st.error("Wpisz pytanie.")
    else:
        from openai import OpenAI
        client = OpenAI(api_key=api_key)

        with st.status("Planowanie zapytań...", expanded=True) as status:
            plan = plan_search(question, list(corpus.keys()), client)
            st.write("Plan:", plan.get("strategy", ""))

            status.update(label="Przeszukiwanie korpusu...")
            hits = search_corpus(corpus, plan.get("chain_queries", []))
            st.write(f"Znaleziono {len(hits)} trafień.")

            status.update(label="Synteza odpowiedzi...")
            if hits:
                answer = synthesize_answer(question, hits, client)
            else:
                answer = "Nie znaleziono pasujących fragmentów w korpusie."
            status.update(label="Gotowe", state="complete")

        st.subheader("Odpowiedź")
        st.markdown(answer)

        tab1, tab2 = st.tabs(["Plan wyszukiwania", "Surowe trafienia"])
        with tab1:
            st.json(plan)
        with tab2:
            if hits:
                rows = [
                    {
                        "plik": h["file"],
                        "linia": h["line_number"],
                        "treść": h["matched_line"].strip(),
                        "kontekst": " | ".join(l.strip() for l in h["context"]),
                    }
                    for h in hits
                ]
                st.dataframe(rows, use_container_width=True)
            else:
                st.write("Brak trafień.")