"""Viewer HTTP server: static files + /api/ask endpoint (graph-aware and AI answer). Replaces plain `python -m http.server 8777`. Endpoints: GET /viewer/* /corpus/* static files GET /api/ask?q=&mode=graph|ai mode=graph -> {"results":[{node...,score,hop}...]} keyword+neighbor BFS scoring mode=ai -> {"answer": "...", "results":[...]} BFS context -> claude-cli answer """ import json import os import re import shutil import subprocess import sys import urllib.parse from pathlib import Path from http.server import SimpleHTTPRequestHandler, ThreadingHTTPServer ROOT = Path(__file__).resolve().parent.parent # C:\Projects\LINE_note OUT = ROOT / "graphify-out" # ---- graph data (loaded once) ---- G = json.loads((OUT / "graph.json").read_text(encoding="utf-8")) NODES = {n["id"]: n for n in G["nodes"]} NODE_LIST = G["nodes"] LINKS = G["links"] ADJ = {} for e in LINKS: ADJ.setdefault(e["source"], []).append((e["target"], e.get("relation", ""))) ADJ.setdefault(e["target"], []).append((e["source"], e.get("relation", ""))) LABELS = {} lp = OUT / ".graphify_labels.json" if lp.exists(): LABELS = json.loads(lp.read_text(encoding="utf-8")) STOP = {"the", "a", "an", "to", "of", "in", "for", "and", "or", "how", "what", "is", "are", "my", "i", "on", "with", "ที่", "และ", "หรือ", "เป็น", "อะไร", "ทำ", "อย่างไร", "คือ", "ใน", "ของ"} def tokenize(q): return [t for t in re.findall(r"[a-z0-9]+|[\u0e00-\u0e7f]+", q.lower()) if len(t) > 1 and t not in STOP] def node_score(n, terms): label = (n.get("label") or "").lower() src = (n.get("source_file") or "").lower() auth = (n.get("author") or "").lower() hits = 0 for t in terms: w = 0 if t in label: w = 6 if label.startswith(t) else 5 if t in src: w = max(w, 3) if t in auth: w = max(w, 2) if w: hits += 1 if not hits: return 0, 0 deg = len(ADJ.get(n["id"], [])) direct = sum(6 if t in label else 3 for t in terms if t in label + src + auth) return direct + min(deg, 30) * 0.1 + hits * 2, 0 def seed_and_expand(terms, top=40, max_hops=2): """Score nodes by direct term match, then add neighbor context from BFS.""" scored = [] for n in NODE_LIST: s, hop = node_score(n, terms) if s > 0: scored.append((s, hop, n)) scored.sort(key=lambda x: -x[0]) seeds = [n for _, _, n in scored[:12]] out = [(s, hop, n["id"]) for s, hop, n in scored[:top]] seen = {nid for _, _, nid in out} # hop-1 neighbors of strong seeds: boost related concept nodes for n in seeds: for nb, rel in ADJ.get(n["id"], [])[:20]: if nb in seen or nb not in NODES: continue m = NODES[nb] s, _ = node_score(m, terms) if s > 0: out.append((s * 0.6, 1, nb)) seen.add(nb) out.sort(key=lambda x: -x[0]) return out[:top], seeds def subgraph_context(terms, budget_chars=48000): """BFS context for LLM: seed docs + their neighbors + relation types.""" top, seeds = seed_and_expand(terms, top=30) lines = [] used = 0 chosen = [] for s, hop, nid in top: n = NODES[nid] chosen.append(nid) line = f"- [{n.get('file_type', '?')}] {n.get('label', '')}" if n.get("author"): line += f" (author: {n['author']})" if n.get("source_url"): line += f" URL: {n['source_url']}" lines.append(line) used += len(line) + 4 if used > budget_chars: break # relations among chosen nodes ids = set(chosen) rels = [] for e in LINKS: if e["source"] in ids and e["target"] in ids: rels.append(f"{NODES[e['source']].get('label','?')} --[{e.get('relation','?')}]--> {NODES[e['target']].get('label','?')}") if len(rels) >= 60: break return "\n".join(lines), rels, chosen AI_PROMPT = """You answer questions about the user's saved knowledge corpus (LINE Note+ links: \ YouTube videos, Facebook posts, articles - Thai and English). Question: {question} Below are the most relevant nodes from the user's knowledge graph (documents and concepts), \ followed by relationships between them. {context} RELATIONSHIPS: {rels} Answer the question in {lang}. Ground your answer in the nodes above: name the specific \ videos/posts/concepts that matter and why. If the corpus contains little on the topic, say \ so plainly. Keep it under 250 words. Structure: short answer first, then supporting nodes.\ """ def lang_of(q): if re.search(r"[\u0e00-\u0e7f]", q): return "Thai" return "the same language as the question" def ask_ai(question, terms): context, rels, chosen = subgraph_context(terms) prompt = AI_PROMPT.format(question=question, context=context, rels="\n".join(rels) or "(none)", lang=lang_of(question)) claude = shutil.which("claude") or shutil.which("claude.exe") if not claude: return "AI backend not configured on this server — only works on the local machine.", chosen cmd = [claude, "-p", prompt, "--model", "sonnet", "--strict-mcp-config", "--max-turns", "1", "--output-format", "text"] try: r = subprocess.run(cmd, capture_output=True, text=True, timeout=180, cwd=str(ROOT), encoding="utf-8", errors="replace") answer = r.stdout.strip() if not answer: answer = "AI backend returned empty answer." + (r.stderr[-300:] if r.stderr else "") except subprocess.TimeoutExpired: answer = "AI backend timed out (180s). Try again." return answer, chosen class Handler(SimpleHTTPRequestHandler): def __init__(self, *a, **kw): super().__init__(*a, directory=str(ROOT), **kw) def do_GET(self): parsed = urllib.parse.urlparse(self.path) if parsed.path == "/api/ask": qs = urllib.parse.parse_qs(parsed.query) q = (qs.get("q") or [""])[0] mode = (qs.get("mode") or ["graph"])[0] if not q.strip(): self._json(400, {"error": "empty query"}) return terms = tokenize(q) if not terms: self._json(400, {"error": "no usable terms"}) return if mode == "ai": answer, chosen = ask_ai(q, terms) self._json(200, {"answer": answer, "ids": chosen}) return top, seeds = seed_and_expand(terms) results = [] for s, hop, nid in top: n = NODES[nid] results.append({ "id": nid, "label": n.get("label"), "score": round(s, 2), "type": n.get("file_type"), "community": n.get("community"), "author": n.get("author"), "url": n.get("source_url"), "source_file": n.get("source_file"), "degree": len(ADJ.get(nid, [])), "neighbors": [NODES[nb]["label"] for nb, _ in ADJ.get(nid, [])[:6] if nb in NODES], }) self._json(200, {"results": results}) return super().do_GET() def _json(self, code, obj): body = json.dumps(obj, ensure_ascii=False).encode("utf-8") self.send_response(code) self.send_header("Content-Type", "application/json; charset=utf-8") self.send_header("Content-Length", str(len(body))) self.send_header("Cache-Control", "no-store") self.end_headers() self.wfile.write(body) def log_message(self, fmt, *args): sys.stderr.write("%s - %s\n" % (self.address_string(), fmt % args)) if __name__ == "__main__": port = int(sys.argv[1]) if len(sys.argv) > 1 else 8777 host = "0.0.0.0" if os.environ.get("VIEWER_BIND_ALL") else "127.0.0.1" srv = ThreadingHTTPServer((host, port), Handler) print(f"viewer server on http://127.0.0.1:{port}", flush=True) srv.serve_forever()