Feat: Discover zukunftssicher (Recency-Score) + Agent-Hirn (Hermes) im Modell-Manager

- discover.rank_runnable: Score aus Fit + Recency (lastModified, Halbwertszeit ~9 Mon)
  + Capability (params, log) + Popularity (downloads, log) -> neuere Generationen bevorzugt.
- Agent-Hirn: neuer GET /api/agent/brain (aktuelles hermes-Modell + bestes NousResearch-
  Hermes-Update, versions-aware via Hermes-X.Y-Parsing). Cockpit zeigt "Agent-Hirn (Hermes)"-Karte
  mit aktuellem Brain + Aktualisieren-Button, wenn NousResearch eine neuere Generation hat
  (z.B. Hermes-4-14B -> Hermes-4.3-36B). Update installiert mit Rolle hermes.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

backend/services/agent.py

@@ -6,14 +6,78 @@ Hermes' eigener Config verdrahtet (siehe docs/HERMES_SETUP.md).
 
 import logging
 import os
+import re
 
 import httpx
+import psutil
 
 from config import ANYTHINGLLM_URL, HERMES_API_URL, HERMES_HOME, PC_EXECUTOR_URL
 
 log = logging.getLogger(__name__)
 
 
+def _hermes_version(name: str) -> float | None:
+    """Versionszahl aus 'Hermes-4.3', 'Hermes-4', 'Nous-Hermes-2' → 4.3/4.0/2.0."""
+    low = (name or "").lower()
+    if "hermes" not in low:
+        return None
+    m = re.search(r"hermes[-_ ]?(\d+(?:\.\d+)?)", low)
+    return float(m.group(1)) if m else None
+
+
+def hermes_brain_info() -> dict:
+    """Aktuelles Agent-Hirn (hermes-Rolle) + bestes verfügbares NousResearch-Hermes-Modell,
+    das auf diese Hardware passt. Für den Modell-Manager: Brain sichtbar + updatebar,
+    sobald NousResearch eine neuere Hermes-Generation veröffentlicht."""
+    from services import discover, llamaswap
+    from services.fit import evaluate_fit, extract_params_b
+
+    models = llamaswap.list_models()
+    cur = next((m for m in models if m.get("role") == "hermes"), None)
+    cur_ver = _hermes_version(cur["name"]) if cur else None
+    cur_params = (cur.get("capabilities") or {}).get("params_b") if cur else None
+    current = None
+    if cur:
+        current = {"name": cur["name"], "filename": cur.get("filename"),
+                   "params_b": cur_params, "quant": cur.get("quant"),
+                   "size_bytes": cur.get("size_bytes"), "version": cur_ver,
+                   "gguf_path": cur.get("gguf_path"), "incomplete": cur.get("incomplete")}
+
+    ram = psutil.virtual_memory().total / (1024 ** 3)
+    best = None
+    try:
+        cands = []
+        for r in discover._fetch_author_models("NousResearch"):
+            rid = r.get("id", "")
+            if "hermes" not in rid.lower():
+                continue
+            pb = extract_params_b(rid)
+            fit = evaluate_fit(pb, "Q4_K_M", 8192, ram, name=rid)
+            if fit["level"] == "too_tight":
+                continue
+            cands.append({"repo": rid, "name": rid.split("/")[-1],
+                          "version": _hermes_version(rid) or 0.0, "params_b": pb,
+                          "downloads": int(r.get("downloads") or 0), "fit": fit})
+        # neueste Hermes-Version zuerst, dann größer/fähiger, dann beliebter
+        cands.sort(key=lambda c: (c["version"], c["params_b"], c["downloads"]), reverse=True)
+        best = cands[0] if cands else None
+    except Exception:
+        log.debug("hermes_brain_info: HF-Abfrage fehlgeschlagen", exc_info=True)
+
+    update = False
+    if best is not None:
+        if cur_ver is None:
+            update = True
+        elif best["version"] > cur_ver:
+            update = True
+        elif best["version"] == cur_ver and best["params_b"] > (cur_params or 0) * 1.05:
+            update = True
+        # gleiche Datei schon installiert? dann kein Update
+        if current and best["repo"].split("/")[-1].lower() in (current["name"] or "").lower():
+            update = False
+    return {"current": current, "recommended": best, "update_available": update}
+
+
 def _reach(url: str, path: str = "") -> bool:
     try:
         with httpx.Client(timeout=3.0) as c: