Feat: Brain-Rolle (hermes) als erstklassige, dauer-warme Hirn-Rolle

- ROLE_IDS (llamaswap, sources) + UI-Rollenlisten (ModelBadges, Discover, ModelBrowse, Cockpit-Slot-Grid) um `hermes` erweitert: gemma erscheint als „Hirn", nicht mehr als scout. - Neuer set_ttl-Helper; set_agent_brain erzwingt ttl:0 (neu + idempotent beim Re-Setzen) und liefert eine weiche Budget-Warnung (kein Hard-Block) bei OOM. - brain_status/brain_model_name: zeigen das echte Hirn (Rolle hermes / Hermes model.default) statt hart `fast` (Bugfix Health-/Ready-Check). - POST /api/models/{id}/role delegiert die Rolle `hermes` an den warm-bewussten Flow (Alias + brains-Gruppe + ttl 0 + Hermes-Config + Gateway-Restart). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-30 17:01:26 +02:00
parent 087eb2259d
commit dd99401f3e
11 changed files with 360 additions and 284 deletions
@@ -7,7 +7,7 @@
    <link rel="manifest" href="/manifest.webmanifest" />
    <link rel="icon" type="image/svg+xml" href="/favicon.svg" />
    <title>Mission Control 2.0</title>
-    <script type="module" crossorigin src="/assets/index-B3e8J_PY.js"></script>
+    <script type="module" crossorigin src="/assets/index-CLiy0fO1.js"></script>
    <link rel="stylesheet" crossorigin href="/assets/index-DoqQd4Zi.css">
  </head>
  <body>
@@ -1,8 +1,9 @@
 import { type Fit } from "@/lib/api"
 import { cn } from "@/lib/utils"

-// Die 5 kanonischen Serving-Rollen (identisch zu Backend sources.py/llamaswap.ROLE_IDS).
-export const ROLES = ["fast", "heavy", "coder", "vision", "scout"]
+// Die kanonischen Serving-Rollen (identisch zu Backend sources.py/llamaswap.ROLE_IDS).
+// `hermes` = Lucys Agent-Hirn (warm + ko-resident); UI-Label „Hirn".
+export const ROLES = ["fast", "heavy", "coder", "vision", "hermes", "scout"]

 // Rolle → Farbton. EINE Quelle der Wahrheit für alle Rollen-Badges (Cockpit, ModelsCard,
 // …). `hermes` = Agent-Hirn (taucht als geladenes Modell auf).
@@ -6,7 +6,7 @@ import { useDialog } from "@/lib/useDialog"
 import { CapsChips } from "@/components/CapsChips"
 import { cn } from "@/lib/utils"
 import { fmtSize, fmtCtx } from "@/lib/format"
-import { getBrandInfo, roleTone } from "@/components/models/ModelBadges"
+import { getBrandInfo, roleTone, ROLES } from "@/components/models/ModelBadges"
 import { SpecDraftModal } from "@/components/models/SpecDraftModal"
 import { LaneEditor } from "@/components/models/LaneEditor"

@@ -107,11 +107,14 @@ export function Cockpit() {

  async function handleRoleChange(role: string, modelName: string) {
    try {
-      await api(`/api/models/${encodeURIComponent(modelName)}/role`, {
+      // Bei der Hirn-Rolle (hermes) läuft serverseitig der warm-bewusste Flow (Alias + brains-
+      // Gruppe + ttl 0 + Hermes-Config + Gateway-Restart) und liefert ggf. eine Budget-Warnung.
+      const res = await api<{ warning?: string | null }>(`/api/models/${encodeURIComponent(modelName)}/role`, {
        method: "POST",
        body: JSON.stringify({ role: role || null }),
      })
      reload()
+      if (res?.warning) showAlert("Hirn gesetzt — Hinweis", res.warning)
    } catch (e: any) {
      showAlert("Fehler", `Fehler beim Zuweisen der Rolle: ${e.message || e}`)
    }
@@ -311,7 +314,7 @@ export function Cockpit() {
          Gateway Steckplatz-Belegung (Slot-Zuweisung)
        </div>
        <div className="grid grid-cols-2 sm:grid-cols-3 lg:grid-cols-6 gap-3">
-          {["fast", "heavy", "coder", "vision", "scout"].map((role) => {
+          {ROLES.map((role) => {
            const m = models.find((x) => x.role === role)
            const isWarm = m ? running.includes(m.name) : false

@@ -1,5 +1,5 @@
 import { useState } from "react"
-import { Download, Star, Layers, Check, Zap, Eye, Code, Brain, Compass, ChevronDown, ChevronUp } from "lucide-react"
+import { Download, Star, Layers, Check, Zap, Eye, Code, Brain, BrainCircuit, Compass, ChevronDown, ChevronUp } from "lucide-react"
 import { api } from "@/lib/api"
 import { useModels, useUpdates, useDiscover } from "@/lib/queries"
 import { cn } from "@/lib/utils"
@@ -7,7 +7,7 @@ import { fmtBytes } from "@/lib/format"
 import { FitBadge } from "@/components/models/ModelBadges"
 import { ModelBrowse } from "./ModelBrowse"

-// Die 5 kanonischen Rollen (identisch zu sources.py / ModelBadges.ROLES).
+// Die kanonischen Rollen (identisch zu sources.py / ModelBadges.ROLES).
 const ROLE_METADATA: Record<string, { title: string; desc: string; icon: any }> = {
  fast: {
    title: "Schnelles Alltags-Hirn",
@@ -29,6 +29,11 @@ const ROLE_METADATA: Record<string, { title: string; desc: string; icon: any }>
    desc: "Verarbeitet Bilder, Screenshots und visuelle Diagramme in multimodalen Chats.",
    icon: Eye
  },
+  hermes: {
+    title: "Lucys Hirn (Agent)",
+    desc: "Lucys dauer-warmes Agent-Hirn mit Tool-Calling — bleibt ko-resident, lädt nie kalt nach.",
+    icon: BrainCircuit
+  },
  scout: {
    title: "Multimodal-Allrounder",
    desc: "Multimodaler Allrounder für schnelle Antworten, Übersetzungen und Alltag.",
@@ -13,7 +13,7 @@ const CHIPS: { key: string; label: string; q: string }[] = [
  { key: "reasoning", label: "Reasoning", q: "reasoning" },
  { key: "small", label: "Klein (≤4B)", q: "3B" },
 ]
-const ROLE_OPTS = ["fast", "heavy", "coder", "vision", "scout"]
+const ROLE_OPTS = ["fast", "heavy", "coder", "vision", "hermes", "scout"]

 const _nf = new Intl.NumberFormat(undefined, { notation: "compact", maximumFractionDigits: 1 })
 function fmtN(n: number): string {