Agent 2.0 exdev 서버 배포 스택

- server-dev start/stop/deploy 및 Gitea push 자동 배포 - local-dev 로컬 개발 환경 Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-21 22:57:30 +09:00
commit 4b86b2a660
344 changed files with 45787 additions and 0 deletions
@@ -0,0 +1,582 @@
+"""
+admin_service.py
+────────────────────────────────────────────
+벡터 DB(Qdrant) 큐레이션 어드민 API + 단일 페이지 웹 UI.
+
+기능:
+  - 검색(semantic): 질문 임베딩 → 벡터 검색
+  - 목록(browse): 페이지네이션 조회
+  - 단건 조회 / 삭제 / 일괄 삭제
+  - 추가: 질문/답변 텍스트 → 임베딩 → 결정적 ID로 upsert
+
+실시간 /ask 서비스(run_service_qa.py)와 같은 모듈/벡터스토어를 공유하되,
+별도 프로세스(컨테이너)로 분리해 운영한다. Qdrant 전용 기능을 사용한다.
+"""
+
+import os
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Optional, List, Dict, Any
+from urllib.parse import unquote
+
+import numpy as np
+from fastapi import FastAPI, HTTPException
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import HTMLResponse
+from pydantic import BaseModel, Field
+
+from api_clients import TEIEmbeddingClient
+from vector_store import get_vector_store, VECTOR_STORE, make_point_id
+
+app = FastAPI(title="RAG 벡터DB 큐레이션 어드민")
+
+# 별도 origin(다른 포트/도메인)의 프론트엔드가 호출할 수 있도록 CORS 허용.
+# 내부망 도구이므로 기본 전체 허용. 필요 시 ADMIN_CORS_ORIGINS="http://host:port,..." 로 제한.
+_cors = os.getenv("ADMIN_CORS_ORIGINS", "*")
+_origins = ["*"] if _cors.strip() == "*" else [o.strip() for o in _cors.split(",") if o.strip()]
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=_origins,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+DATA_DIR = Path(os.getenv("DATA_DIR", "/app/data"))
+
+
+def _now_iso() -> str:
+    """등록/수정 시각 기본값 (UTC ISO-8601)"""
+    return datetime.now(timezone.utc).isoformat()
+
+print("[Admin] 초기화 시작...")
+embed_client = TEIEmbeddingClient()
+vector_store = get_vector_store()
+vector_store.load(str(DATA_DIR))
+print(f"[Admin] 벡터 스토어({VECTOR_STORE}) 로드 완료: {vector_store.count()}개")
+
+if VECTOR_STORE != "qdrant":
+    print("[Admin] ⚠️ 경고: 어드민의 조회/삭제/추가 기능은 Qdrant에서만 동작합니다 "
+          f"(현재 VECTOR_STORE={VECTOR_STORE}).")
+else:
+    vector_store.ensure_admin_payload_indexes()
+
+
+# ───────────────────────────────────────────
+# 요청/응답 모델
+# ───────────────────────────────────────────
+class SearchRequest(BaseModel):
+    query: str = Field(..., min_length=1, max_length=500)
+    top_k: int = Field(20, ge=1, le=100)
+    threshold: Optional[float] = None
+    category: Optional[str] = None   # 정확일치 필터
+    source: Optional[str] = None     # 정확일치 필터
+
+
+class AddRequest(BaseModel):
+    q: str = Field(..., min_length=1)
+    a: str = Field(..., min_length=1)
+    category: Optional[str] = None
+    url: Optional[str] = None
+    source: str = "admin_manual"
+    source_id: Optional[str] = None
+    source_created_at: Optional[str] = None
+
+
+class UpdateRequest(BaseModel):
+    """전달된 필드만 수정. q가 바뀌면 재임베딩한다."""
+    q: Optional[str] = None
+    a: Optional[str] = None
+    category: Optional[str] = None
+    url: Optional[str] = None
+    source_created_at: Optional[str] = None
+
+
+class KeywordSearchRequest(BaseModel):
+    """DB 전체 대상 문자열 포함(부분일치) 검색"""
+    keyword: str = Field(..., min_length=1, max_length=200)
+    field: str = Field("both", pattern="^(both|q|a)$")  # 검색 대상 필드
+    category: Optional[str] = None
+    source: Optional[str] = None
+    skip: int = Field(0, ge=0)
+    limit: int = Field(50, ge=1, le=200)
+
+
+class DeleteBatchRequest(BaseModel):
+    ids: List[str] = Field(..., min_items=1)
+
+
+def _row(item: Dict[str, Any], score: Optional[float] = None) -> Dict[str, Any]:
+    """검색/조회 결과를 UI 친화 형태로 변환"""
+    meta = item.get("meta", {}) or {}
+    row = {
+        "id": item.get("id"),
+        "q": meta.get("q"),
+        "a": meta.get("a"),
+        "category": meta.get("category"),
+        "source": meta.get("source"),
+        "source_id": meta.get("source_id"),
+        "url": meta.get("url"),
+        "source_created_at": meta.get("source_created_at"),
+        "indexed_at": meta.get("indexed_at"),
+        "updated_at": meta.get("updated_at"),
+    }
+    if score is not None:
+        row["score"] = round(float(score), 4)
+    return row
+
+
+def _require_qdrant():
+    if VECTOR_STORE != "qdrant":
+        raise HTTPException(
+            status_code=400,
+            detail="이 기능은 Qdrant 벡터스토어에서만 지원됩니다. VECTOR_STORE=qdrant 로 실행하세요.",
+        )
+
+
+# ───────────────────────────────────────────
+# API
+# ───────────────────────────────────────────
+@app.get("/api/stats")
+def stats():
+    return {"vector_store": VECTOR_STORE, "count": vector_store.count()}
+
+
+def _normalize_query_param(value: Optional[str]) -> Optional[str]:
+    """쿼리 파라미터 이중 URL 인코딩 복구 (Java RestTemplate + toUriString 조합 대응)."""
+    if value is None:
+        return None
+    v = value.strip()
+    if not v:
+        return v
+    for _ in range(3):
+        decoded = unquote(v)
+        if decoded == v:
+            break
+        v = decoded
+    return v
+
+
+def _filter_dict(category: Optional[str], source: Optional[str]) -> Optional[Dict[str, Any]]:
+    f: Dict[str, Any] = {}
+    if category:
+        cat = (_normalize_query_param(category) or category).strip()
+        if cat == "미분류":
+            f["category"] = "__EMPTY__"
+        else:
+            f["category"] = cat
+    if source:
+        src = (_normalize_query_param(source) or source).strip()
+        if src:
+            f["source"] = src
+    return f or None
+
+
+@app.post("/api/search")
+def search(req: SearchRequest):
+    vecs = embed_client.embed([req.query], normalize=True, is_query=True)
+    if not vecs:
+        raise HTTPException(status_code=502, detail="임베딩 실패")
+    query_vec = np.array(vecs[0], dtype="float32")
+    results = vector_store.search(
+        query_vec,
+        top_k=req.top_k,
+        threshold=req.threshold,
+        filter_dict=_filter_dict(req.category, req.source),
+    )
+    return {"count": len(results), "items": [_row(r, r.get("score")) for r in results]}
+
+
+@app.post("/api/keyword-search")
+def keyword_search(req: KeywordSearchRequest):
+    """문자열 포함(부분일치) 검색 — 의미 검색과 달리 키워드가 실제 포함된 항목만 반환"""
+    _require_qdrant()
+    fields = {"both": ("q", "a"), "q": ("q",), "a": ("a",)}[req.field]
+    items, total, exhausted = vector_store.keyword_search(
+        req.keyword,
+        fields=fields,
+        filter_dict=_filter_dict(req.category, req.source),
+        skip=req.skip,
+        limit=req.limit,
+    )
+    return {
+        "count": len(items),
+        "total": total,
+        "exhausted": exhausted,
+        "items": [_row(it) for it in items],
+    }
+
+
+@app.get("/api/categories")
+def categories():
+    """등록된 category 목록 + 건수 (어드민 분류 필터/트리용)"""
+    _require_qdrant()
+    counts = vector_store.distinct_payload_values("category")
+    items = [{"category": k, "count": v} for k, v in sorted(counts.items())]
+    return {"count": len(items), "items": items}
+
+
+@app.get("/api/points")
+def list_points(
+    limit: int = 50,
+    offset: Optional[str] = None,
+    page: Optional[int] = None,
+    size: Optional[int] = None,
+    category: Optional[str] = None,
+    source: Optional[str] = None,
+):
+    _require_qdrant()
+    fd = _filter_dict(category, source)
+    if page is not None:
+        pg = max(0, page)
+        sz = max(1, min(size or 20, 200))
+        items, total = vector_store.list_points_page(pg, sz, fd)
+        return {
+            "page": pg,
+            "size": sz,
+            "total": total,
+            "items": [_row(it) for it in items],
+        }
+    items, next_offset = vector_store.list_points(
+        limit=limit, offset=offset, filter_dict=fd
+    )
+    return {
+        "count": len(items),
+        "items": [_row(it) for it in items],
+        "next_offset": next_offset,
+    }
+
+
+@app.get("/api/points/{point_id}")
+def get_point(point_id: str):
+    _require_qdrant()
+    item = vector_store.get_by_id(point_id)
+    if not item:
+        raise HTTPException(status_code=404, detail="해당 ID의 항목이 없습니다.")
+    return _row(item)
+
+
+@app.put("/api/points/{point_id}")
+def update_point(point_id: str, req: UpdateRequest):
+    """
+    기존 항목 수정. 전달된 필드만 갱신한다.
+    - q 기반 결정적 ID → q 변경 시 새 ID로 upsert 후 옛 ID 삭제
+    - 추가/업로드와 동일: 더 최신 source_created_at이 이미 있으면 스킵
+    """
+    _require_qdrant()
+    existing = vector_store.get_by_id(point_id)
+    if not existing:
+        raise HTTPException(status_code=404, detail="해당 ID의 항목이 없습니다.")
+
+    meta: Dict[str, Any] = dict(existing.get("meta") or {})
+    if req.q is not None:
+        meta["q"] = req.q.strip()
+    if req.a is not None:
+        meta["a"] = req.a
+    if req.category is not None:
+        meta["category"] = req.category
+    if req.url is not None:
+        meta["url"] = req.url
+    if req.source_created_at is not None:
+        meta["source_created_at"] = req.source_created_at.strip() or None
+
+    if not meta.get("q") or not meta.get("a"):
+        raise HTTPException(status_code=400, detail="q와 a는 비울 수 없습니다.")
+    if not meta.get("source_created_at") or not str(meta.get("source_created_at")).strip():
+        raise HTTPException(status_code=400, detail="source_created_at은 필수입니다.")
+    meta["source_created_at"] = str(meta["source_created_at"]).strip()
+
+    # 질문 재임베딩 (q가 안 바뀌었어도 일관성을 위해 항상 재임베딩)
+    vecs = embed_client.embed([meta["q"]], normalize=True, is_query=False)
+    if not vecs:
+        raise HTTPException(status_code=502, detail="임베딩 실패")
+
+    new_id = make_point_id(meta)
+    vectors = np.array([vecs[0]], dtype="float32")
+
+    if hasattr(vector_store, "upsert_vectors"):
+        stats = vector_store.upsert_vectors(vectors, [meta], skip_if_older=True)
+        if stats.get("skipped"):
+            return {
+                "skipped": True,
+                "reason": stats.get("last_skip_reason") or "older_source_created_at",
+                "id": stats.get("last_id") or new_id,
+            }
+    else:
+        meta.setdefault("indexed_at", _now_iso())
+        meta["updated_at"] = _now_iso()
+        vector_store.add_vectors(vectors, [meta], skip_if_older=True)
+
+    # ID가 바뀐 경우(질문 변경) 옛 항목 제거
+    if new_id != point_id:
+        vector_store.delete_by_id(point_id)
+    # 레거시(source_id 기반) ID로 남아 있는 동일 q 항목 정리
+    if hasattr(vector_store, "find_by_exact_q"):
+        legacy = vector_store.find_by_exact_q(meta["q"])
+        if legacy and str(legacy.get("id")) not in (str(new_id), str(point_id)):
+            vector_store.delete_by_id(str(legacy["id"]))
+
+    return {"updated": new_id, "moved": new_id != point_id, "skipped": False}
+
+
+@app.delete("/api/points/{point_id}")
+def delete_point(point_id: str):
+    _require_qdrant()
+    if not vector_store.get_by_id(point_id):
+        raise HTTPException(status_code=404, detail="해당 ID의 항목이 없습니다.")
+    vector_store.delete_by_id(point_id)
+    return {"deleted": point_id}
+
+
+@app.post("/api/points/delete-batch")
+def delete_batch(req: DeleteBatchRequest):
+    _require_qdrant()
+    vector_store.delete_by_ids(req.ids)
+    return {"deleted": len(req.ids)}
+
+
+@app.post("/api/points")
+def add_point(req: AddRequest):
+    _require_qdrant()
+    if not req.source_created_at or not req.source_created_at.strip():
+        raise HTTPException(status_code=400, detail="source_created_at은 필수입니다.")
+
+    # 문서 임베딩(is_query=False) → ingest와 동일 방식
+    vecs = embed_client.embed([req.q], normalize=True, is_query=False)
+    if not vecs:
+        raise HTTPException(status_code=502, detail="임베딩 실패")
+
+    source_id = req.source_id or f"manual_{os.urandom(6).hex()}"
+    now = _now_iso()
+    meta: Dict[str, Any] = {
+        "q": req.q.strip(),
+        "a": req.a,
+        "category": req.category,
+        "source": req.source,
+        "source_id": source_id,
+        "url": req.url,
+        "source_created_at": req.source_created_at.strip(),
+        "indexed_at": now,
+        "updated_at": now,
+    }
+    point_id = make_point_id(meta)
+    vectors = np.array([vecs[0]], dtype="float32")
+
+    if hasattr(vector_store, "upsert_vectors"):
+        stats = vector_store.upsert_vectors(vectors, [meta], skip_if_older=True)
+        if stats.get("skipped"):
+            return {
+                "skipped": True,
+                "reason": stats.get("last_skip_reason") or "older_source_created_at",
+                "id": stats.get("last_id") or point_id,
+            }
+        action = stats.get("last_action") or "inserted"
+        return {
+            "created": stats.get("last_id") or point_id,
+            "source_id": source_id,
+            "action": action,
+            "skipped": False,
+        }
+
+    vector_store.add_vectors(vectors, [meta], skip_if_older=True)
+    return {"created": point_id, "source_id": source_id, "skipped": False}
+
+
+# ───────────────────────────────────────────
+# 웹 UI (빌드 불필요 단일 페이지)
+# ───────────────────────────────────────────
+@app.get("/", response_class=HTMLResponse)
+def index():
+    return HTML_PAGE
+
+
+HTML_PAGE = """<!doctype html>
+<html lang="ko">
+<head>
+<meta charset="utf-8"/>
+<meta name="viewport" content="width=device-width, initial-scale=1"/>
+<title>벡터DB 큐레이션 어드민</title>
+<style>
+  :root { --bd:#e2e8f0; --pri:#2563eb; --bg:#f8fafc; --txt:#0f172a; --muted:#64748b; --danger:#dc2626; }
+  * { box-sizing:border-box; }
+  body { margin:0; font-family:'Segoe UI',-apple-system,system-ui,sans-serif; background:var(--bg); color:var(--txt); }
+  header { background:#fff; border-bottom:1px solid var(--bd); padding:16px 24px; display:flex; align-items:center; gap:16px; position:sticky; top:0; z-index:10; }
+  header h1 { font-size:18px; margin:0; }
+  .stat { margin-left:auto; color:var(--muted); font-size:14px; }
+  main { max-width:1100px; margin:0 auto; padding:24px; }
+  .tabs { display:flex; gap:8px; margin-bottom:16px; }
+  .tab { padding:8px 16px; border:1px solid var(--bd); border-radius:8px; background:#fff; cursor:pointer; font-size:14px; }
+  .tab.active { background:var(--pri); color:#fff; border-color:var(--pri); }
+  .panel { display:none; background:#fff; border:1px solid var(--bd); border-radius:12px; padding:20px; }
+  .panel.active { display:block; }
+  .row { display:flex; gap:8px; align-items:center; flex-wrap:wrap; }
+  input, textarea, select { width:100%; padding:9px 12px; border:1px solid var(--bd); border-radius:8px; font-size:14px; font-family:inherit; }
+  textarea { min-height:90px; resize:vertical; }
+  label { display:block; font-size:13px; color:var(--muted); margin:10px 0 4px; }
+  button { padding:9px 16px; border:none; border-radius:8px; background:var(--pri); color:#fff; cursor:pointer; font-size:14px; }
+  button.secondary { background:#fff; color:var(--txt); border:1px solid var(--bd); }
+  button.danger { background:var(--danger); }
+  button:disabled { opacity:.5; cursor:not-allowed; }
+  .card { border:1px solid var(--bd); border-radius:10px; padding:14px; margin-top:12px; }
+  .card .meta { font-size:12px; color:var(--muted); margin-bottom:6px; display:flex; gap:10px; flex-wrap:wrap; }
+  .card .q { font-weight:600; margin-bottom:6px; }
+  .card .a { font-size:14px; color:#334155; white-space:pre-wrap; }
+  .badge { background:#eff6ff; color:var(--pri); border-radius:6px; padding:1px 8px; font-size:12px; }
+  .toolbar { display:flex; gap:8px; margin-top:10px; }
+  .empty { color:var(--muted); text-align:center; padding:30px; }
+  .toast { position:fixed; bottom:20px; left:50%; transform:translateX(-50%); background:#0f172a; color:#fff; padding:10px 18px; border-radius:8px; opacity:0; transition:.2s; font-size:14px; }
+  .toast.show { opacity:1; }
+  code { background:#f1f5f9; padding:1px 6px; border-radius:4px; font-size:12px; }
+</style>
+</head>
+<body>
+<header>
+  <h1>🗂️ 벡터DB 큐레이션 어드민</h1>
+  <div class="stat" id="stat">로딩...</div>
+</header>
+<main>
+  <div class="tabs">
+    <div class="tab active" data-tab="search">의미검색</div>
+    <div class="tab" data-tab="keyword">키워드검색</div>
+    <div class="tab" data-tab="browse">목록</div>
+    <div class="tab" data-tab="add">추가</div>
+  </div>
+
+  <!-- 의미 검색 -->
+  <section class="panel active" id="panel-search">
+    <div class="row">
+      <input id="q" placeholder="질문을 입력하세요 (뜻이 비슷한 항목 검색 = 챗봇과 동일)" onkeydown="if(event.key==='Enter')doSearch()"/>
+      <button onclick="doSearch()">검색</button>
+    </div>
+    <div id="search-results"></div>
+  </section>
+
+  <!-- 키워드(문자열 포함) 검색 -->
+  <section class="panel" id="panel-keyword">
+    <div class="row">
+      <input id="kw" placeholder="포함된 문자열 검색 (예: 1588-2504, 부가통행료)" onkeydown="if(event.key==='Enter')doKeyword()"/>
+      <select id="kw-field" style="max-width:140px">
+        <option value="both">질문+답변</option>
+        <option value="q">질문만</option>
+        <option value="a">답변만</option>
+      </select>
+      <button onclick="doKeyword()">검색</button>
+    </div>
+    <div id="kw-results"></div>
+  </section>
+
+  <!-- 목록 -->
+  <section class="panel" id="panel-browse">
+    <div class="row">
+      <button class="secondary" onclick="loadList(true)">처음부터</button>
+      <button class="secondary" id="more-btn" onclick="loadList(false)">더 보기</button>
+    </div>
+    <div id="list-results"></div>
+  </section>
+
+  <!-- 추가 -->
+  <section class="panel" id="panel-add">
+    <label>질문 (q) *</label>
+    <input id="add-q" placeholder="예: 하이패스 단말기는 어디서 구입하나요?"/>
+    <label>답변 (a) *</label>
+    <textarea id="add-a" placeholder="답변 내용"></textarea>
+    <div class="row">
+      <div style="flex:1"><label>분류 (category)</label><input id="add-cat" placeholder="예: 하이패스 이용"/></div>
+      <div style="flex:1"><label>URL</label><input id="add-url" placeholder="https://..."/></div>
+    </div>
+    <div class="toolbar"><button onclick="doAdd()">벡터DB에 추가</button></div>
+    <p style="color:var(--muted);font-size:13px">추가 시 질문이 임베딩되어 검색 대상이 됩니다. 같은 항목을 다시 추가하면 덮어쓰기됩니다.</p>
+  </section>
+</main>
+<div class="toast" id="toast"></div>
+
+<script>
+const $ = (s) => document.querySelector(s);
+let listOffset = null;
+
+function toast(msg){ const t=$("#toast"); t.textContent=msg; t.classList.add("show"); setTimeout(()=>t.classList.remove("show"),2200); }
+function esc(s){ return (s??"").toString().replace(/[&<>]/g, c=>({'&':'&amp;','<':'&lt;','>':'&gt;'}[c])); }
+
+document.querySelectorAll(".tab").forEach(t=>t.onclick=()=>{
+  document.querySelectorAll(".tab").forEach(x=>x.classList.remove("active"));
+  document.querySelectorAll(".panel").forEach(x=>x.classList.remove("active"));
+  t.classList.add("active");
+  $("#panel-"+t.dataset.tab).classList.add("active");
+  if(t.dataset.tab==="browse") loadList(true);
+});
+
+async function refreshStat(){
+  try{ const r=await fetch("api/stats"); const d=await r.json(); $("#stat").textContent=`벡터 ${d.count.toLocaleString()}개 · ${d.vector_store}`; }
+  catch(e){ $("#stat").textContent="상태 조회 실패"; }
+}
+
+function cardHtml(it, showScore){
+  const score = showScore && it.score!=null ? `<span class="badge">score ${it.score}</span>` : "";
+  return `<div class="card" data-id="${esc(it.id)}">
+    <div class="meta">${score}<span>출처: ${esc(it.source)||'-'}</span><span>분류: ${esc(it.category)||'-'}</span><span>id: <code>${esc(it.id)}</code></span></div>
+    <div class="q">Q. ${esc(it.q)}</div>
+    <div class="a">A. ${esc(it.a)}</div>
+    ${it.url?`<div class="meta"><a href="${esc(it.url)}" target="_blank">${esc(it.url)}</a></div>`:""}
+    <div class="toolbar"><button class="danger" onclick="del('${esc(it.id)}')">삭제</button></div>
+  </div>`;
+}
+
+async function doSearch(){
+  const q=$("#q").value.trim(); if(!q) return;
+  $("#search-results").innerHTML="<div class='empty'>검색 중...</div>";
+  try{
+    const r=await fetch("api/search",{method:"POST",headers:{"Content-Type":"application/json"},body:JSON.stringify({query:q,top_k:20})});
+    const d=await r.json();
+    $("#search-results").innerHTML = d.items.length ? d.items.map(it=>cardHtml(it,true)).join("") : "<div class='empty'>결과가 없습니다.</div>";
+  }catch(e){ $("#search-results").innerHTML="<div class='empty'>검색 실패</div>"; }
+}
+
+async function doKeyword(){
+  const kw=$("#kw").value.trim(); if(!kw) return;
+  $("#kw-results").innerHTML="<div class='empty'>검색 중...</div>";
+  try{
+    const r=await fetch("api/keyword-search",{method:"POST",headers:{"Content-Type":"application/json"},
+      body:JSON.stringify({keyword:kw,field:$("#kw-field").value,limit:100})});
+    const d=await r.json();
+    const note=`<div class="meta" style="margin-bottom:8px">스캔 ${d.scanned}건 · 매칭 ${d.count}건${d.exhausted?'':' (상위 일부만 표시)'}</div>`;
+    $("#kw-results").innerHTML = note + (d.items.length ? d.items.map(it=>cardHtml(it,false)).join("") : "<div class='empty'>포함된 항목이 없습니다.</div>");
+  }catch(e){ $("#kw-results").innerHTML="<div class='empty'>검색 실패</div>"; }
+}
+
+async function loadList(reset){
+  if(reset){ listOffset=null; $("#list-results").innerHTML=""; }
+  try{
+    const url = "api/points?limit=20" + (listOffset?("&offset="+encodeURIComponent(listOffset)):"");
+    const r=await fetch(url); const d=await r.json();
+    if(reset && !d.items.length){ $("#list-results").innerHTML="<div class='empty'>등록된 항목이 없습니다.</div>"; }
+    else { $("#list-results").insertAdjacentHTML("beforeend", d.items.map(it=>cardHtml(it,false)).join("")); }
+    listOffset = d.next_offset;
+    $("#more-btn").disabled = !listOffset;
+  }catch(e){ toast("목록 조회 실패"); }
+}
+
+async function del(id){
+  if(!confirm("이 항목을 벡터DB에서 삭제할까요?")) return;
+  try{
+    const r=await fetch("api/points/"+encodeURIComponent(id),{method:"DELETE"});
+    if(!r.ok) throw 0;
+    document.querySelector(`.card[data-id="${CSS.escape(id)}"]`)?.remove();
+    toast("삭제되었습니다."); refreshStat();
+  }catch(e){ toast("삭제 실패"); }
+}
+
+async function doAdd(){
+  const q=$("#add-q").value.trim(), a=$("#add-a").value.trim();
+  if(!q||!a){ toast("질문과 답변은 필수입니다."); return; }
+  try{
+    const r=await fetch("api/points",{method:"POST",headers:{"Content-Type":"application/json"},
+      body:JSON.stringify({q,a,category:$("#add-cat").value.trim()||null,url:$("#add-url").value.trim()||null})});
+    if(!r.ok) throw 0;
+    $("#add-q").value=""; $("#add-a").value=""; $("#add-cat").value=""; $("#add-url").value="";
+    toast("추가되었습니다."); refreshStat();
+  }catch(e){ toast("추가 실패"); }
+}
+
+refreshStat();
+</script>
+</body>
+</html>"""