feat(azure-ai+gpu-ui): per-tenant Azure AI config + GPU health panel

- Per-tenant Azure AI config stored in tenants.tenant_config JSONB
- GET/PUT /api/tenants/{id}/ai-config + POST .../test connection
- api_key never returned to frontend (has_api_key: bool pattern)
- azure_ai.py resolves creds from tenant config when ai_enabled=True
- ai_tasks.py loads tenant config and passes it to validate_thumbnail
- Admin GPU Status section: probe button + status badge + last-checked time
- Notifications: _BELL_CHANNELS filter (notification+alert only in bell)
- Tenants.tsx: per-row Azure AI Config modal with URL auto-parse helper
- Remove duplicate in-memory /gpu-probe endpoints (kept DB-backed /probe/gpu)

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

backend/app/api/routers/notifications.py

@@ -51,30 +51,42 @@ def _visibility_filter(user: User):
     return and_(AuditLog.notification == True, targeted)  # noqa: E712
 
 
+# Default channels shown in bell dropdown
+_BELL_CHANNELS = ("notification", "alert")
+
+
 @router.get("", response_model=NotificationListResponse)
 async def list_notifications(
     limit: int = Query(20, ge=1, le=100),
     offset: int = Query(0, ge=0),
     unread_only: bool = Query(False),
+    channel: Optional[str] = Query(None, description="Filter by channel: notification, activity, alert. Defaults to notification+alert."),
     user: User = Depends(get_current_user),
     db: AsyncSession = Depends(get_db),
 ):
     vis = _visibility_filter(user)
 
+    # Channel filter
+    if channel:
+        channel_filter = AuditLog.channel == channel
+    else:
+        channel_filter = AuditLog.channel.in_(_BELL_CHANNELS)
+
     # Total count
-    total_q = select(func.count(AuditLog.id)).where(vis)
+    total_q = select(func.count(AuditLog.id)).where(vis, channel_filter)
     if unread_only:
         total_q = total_q.where(AuditLog.read_at.is_(None))
     total = (await db.execute(total_q)).scalar() or 0
 
-    # Unread count (always)
-    unread_q = select(func.count(AuditLog.id)).where(vis, AuditLog.read_at.is_(None))
+    # Unread count (always — only for bell channels, not activity)
+    bell_channel_filter = AuditLog.channel.in_(_BELL_CHANNELS)
+    unread_q = select(func.count(AuditLog.id)).where(vis, bell_channel_filter, AuditLog.read_at.is_(None))
     unread_count = (await db.execute(unread_q)).scalar() or 0
 
     # Items
     items_q = (
         select(AuditLog)
-        .where(vis)
+        .where(vis, channel_filter)
         .order_by(AuditLog.timestamp.desc())
         .offset(offset)
         .limit(limit)
@@ -105,7 +117,8 @@ async def unread_count(
     db: AsyncSession = Depends(get_db),
 ):
     vis = _visibility_filter(user)
-    q = select(func.count(AuditLog.id)).where(vis, AuditLog.read_at.is_(None))
+    bell_channel_filter = AuditLog.channel.in_(_BELL_CHANNELS)
+    q = select(func.count(AuditLog.id)).where(vis, bell_channel_filter, AuditLog.read_at.is_(None))
     count = (await db.execute(q)).scalar() or 0
     return UnreadCountResponse(unread_count=count)
 
@@ -182,7 +195,9 @@ async def update_my_notification_config(
 ):
     if channel not in ("in_app", "email"):
         raise HTTPException(status_code=400, detail="channel must be 'in_app' or 'email'")
-    return await upsert_notification_config(db, current_user.id, event_type, channel, body.enabled)
+    if body.frequency is not None and body.frequency not in ("immediate", "daily", "never"):
+        raise HTTPException(status_code=400, detail="frequency must be 'immediate', 'daily', or 'never'")
+    return await upsert_notification_config(db, current_user.id, event_type, channel, body.enabled, body.frequency)
 
 
 @router.post("/config/reset", response_model=list[NotificationConfigOut])

backend/app/api/routers/worker.py

@@ -683,3 +683,4 @@ async def update_worker_config(
         enabled=cfg.enabled,
         updated_at=cfg.updated_at.isoformat(),
     )
+

backend/app/domains/tenants/models.py

@@ -12,6 +12,15 @@ _DEFAULT_TENANT_CONFIG = {
     "fallback_material": "SCHAEFFLER_059999_FailedMaterial",
     "notifications_enabled": True,
     "invoice_prefix": "INV",
+    # Azure AI validation (per-tenant)
+    "ai_enabled": False,
+    "ai_api_key": None,          # stored but never returned to frontend
+    "ai_endpoint": None,         # e.g. https://myinstance.openai.azure.com
+    "ai_deployment": "gpt-4o",
+    "ai_api_version": "2024-02-01",
+    "ai_max_tokens": 500,
+    "ai_temperature": 0.1,
+    "ai_validation_prompt": None,  # None = use DEFAULT_VALIDATION_PROMPT from azure_ai.py
 }
 
 

backend/app/domains/tenants/router.py

@@ -1,11 +1,16 @@
 import uuid
 from fastapi import APIRouter, Depends, HTTPException, status
 from sqlalchemy.ext.asyncio import AsyncSession
+from sqlalchemy import update
 
 from app.database import get_db
 from app.utils.auth import require_admin
-from app.domains.tenants.schemas import TenantCreate, TenantUpdate, TenantOut
+from app.domains.tenants.schemas import (
+    TenantCreate, TenantUpdate, TenantOut,
+    TenantAIConfigUpdate, TenantAIConfigOut,
+)
 from app.domains.tenants import service
+from app.domains.tenants.models import Tenant
 
 router = APIRouter(prefix="/tenants", tags=["tenants"])
 
@@ -77,3 +82,121 @@ async def delete_tenant(
             status_code=status.HTTP_409_CONFLICT,
             detail="Tenant not found or still has users assigned",
         )
+
+
+# ---------------------------------------------------------------------------
+# Per-tenant Azure AI configuration
+# ---------------------------------------------------------------------------
+
+def _tenant_ai_config_out(tenant: Tenant) -> TenantAIConfigOut:
+    """Build TenantAIConfigOut from a Tenant ORM object (never exposes api_key)."""
+    cfg = tenant.tenant_config or {}
+    return TenantAIConfigOut(
+        ai_enabled=cfg.get("ai_enabled", False),
+        ai_endpoint=cfg.get("ai_endpoint"),
+        ai_deployment=cfg.get("ai_deployment", "gpt-4o"),
+        ai_api_version=cfg.get("ai_api_version", "2024-02-01"),
+        has_api_key=bool(cfg.get("ai_api_key")),
+        ai_max_tokens=cfg.get("ai_max_tokens", 500),
+        ai_temperature=cfg.get("ai_temperature", 0.1),
+        ai_validation_prompt=cfg.get("ai_validation_prompt"),
+    )
+
+
+@router.get("/{tenant_id}/ai-config", response_model=TenantAIConfigOut)
+async def get_tenant_ai_config(
+    tenant_id: uuid.UUID,
+    db: AsyncSession = Depends(get_db),
+    _: object = Depends(require_admin),
+):
+    """Return AI config for a tenant (without the raw api_key)."""
+    tenant = await service.get_tenant(db, tenant_id)
+    if not tenant:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Tenant not found")
+    return _tenant_ai_config_out(tenant)
+
+
+@router.put("/{tenant_id}/ai-config", response_model=TenantAIConfigOut)
+async def update_tenant_ai_config(
+    tenant_id: uuid.UUID,
+    body: TenantAIConfigUpdate,
+    db: AsyncSession = Depends(get_db),
+    _: object = Depends(require_admin),
+):
+    """Merge AI configuration into tenant_config JSONB.
+    If ai_api_key is None in the request body, the existing key is preserved.
+    """
+    tenant = await service.get_tenant(db, tenant_id)
+    if not tenant:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Tenant not found")
+
+    # Build the merged config dict
+    current_cfg: dict = dict(tenant.tenant_config or {})
+
+    current_cfg["ai_enabled"] = body.ai_enabled
+    current_cfg["ai_endpoint"] = body.ai_endpoint
+    current_cfg["ai_deployment"] = body.ai_deployment
+    current_cfg["ai_api_version"] = body.ai_api_version
+    current_cfg["ai_max_tokens"] = body.ai_max_tokens
+    current_cfg["ai_temperature"] = body.ai_temperature
+    current_cfg["ai_validation_prompt"] = body.ai_validation_prompt
+
+    # Only overwrite the stored key if a new value was actually provided
+    if body.ai_api_key is not None:
+        current_cfg["ai_api_key"] = body.ai_api_key
+
+    # Use direct SQL UPDATE to reliably persist JSONB mutations
+    await db.execute(
+        update(Tenant)
+        .where(Tenant.id == tenant_id)
+        .values(tenant_config=current_cfg)
+    )
+    await db.commit()
+    await db.refresh(tenant)
+    return _tenant_ai_config_out(tenant)
+
+
+@router.post("/{tenant_id}/ai-config/test")
+async def test_tenant_ai_config(
+    tenant_id: uuid.UUID,
+    db: AsyncSession = Depends(get_db),
+    _: object = Depends(require_admin),
+):
+    """Send a minimal ping to Azure OpenAI using the tenant's stored credentials.
+    Returns {"ok": true} or {"ok": false, "error": "human readable message"}.
+    """
+    tenant = await service.get_tenant(db, tenant_id)
+    if not tenant:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Tenant not found")
+
+    cfg = tenant.tenant_config or {}
+    api_key = cfg.get("ai_api_key")
+    endpoint = cfg.get("ai_endpoint")
+    deployment = cfg.get("ai_deployment", "gpt-4o")
+    api_version = cfg.get("ai_api_version", "2024-02-01")
+
+    if not api_key:
+        return {"ok": False, "error": "No API key configured for this tenant"}
+    if not endpoint:
+        return {"ok": False, "error": "No endpoint configured for this tenant"}
+
+    import asyncio
+
+    def _ping() -> dict:
+        try:
+            from openai import AzureOpenAI
+            client = AzureOpenAI(
+                api_key=api_key,
+                azure_endpoint=endpoint,
+                api_version=api_version,
+            )
+            client.chat.completions.create(
+                model=deployment,
+                messages=[{"role": "user", "content": "ping"}],
+                max_tokens=5,
+            )
+            return {"ok": True}
+        except Exception as exc:
+            return {"ok": False, "error": str(exc)}
+
+    return await asyncio.to_thread(_ping)

backend/app/domains/tenants/schemas.py

@@ -24,3 +24,25 @@ class TenantOut(BaseModel):
     created_at: datetime
 
     model_config = {"from_attributes": True}
+
+
+class TenantAIConfigUpdate(BaseModel):
+    ai_enabled: bool = False
+    ai_endpoint: str | None = None
+    ai_deployment: str = "gpt-4o"
+    ai_api_version: str = "2024-02-01"
+    ai_api_key: str | None = None     # optional — don't require re-entry
+    ai_max_tokens: int = 500
+    ai_temperature: float = 0.1
+    ai_validation_prompt: str | None = None
+
+
+class TenantAIConfigOut(BaseModel):
+    ai_enabled: bool
+    ai_endpoint: str | None
+    ai_deployment: str
+    ai_api_version: str
+    has_api_key: bool              # True if ai_api_key is set — never return the key itself
+    ai_max_tokens: int
+    ai_temperature: float
+    ai_validation_prompt: str | None

backend/app/services/azure_ai.py

@@ -24,10 +24,13 @@ Respond in JSON with exactly these fields:
 }"""
 
 
-def validate_thumbnail(order_item_id: str) -> dict:
+def validate_thumbnail(order_item_id: str, tenant_config: dict | None = None) -> dict:
     """
     Validate thumbnail orientation using Azure GPT-4o Vision.
     Updates the order_item AI validation fields in DB.
+
+    If tenant_config is provided and tenant_config["ai_enabled"] is True,
+    the tenant's own Azure credentials are used instead of global settings.
     """
     from app.config import settings
     from sqlalchemy import create_engine
@@ -45,7 +48,7 @@ def validate_thumbnail(order_item_id: str) -> dict:
         session.commit()
 
         try:
-            result = _call_azure_vision(item.thumbnail_path, settings)
+            result = _call_azure_vision(item.thumbnail_path, settings, tenant_config)
             item.ai_validation_status = AIValidationStatus.completed
             item.ai_validation_result = result
         except Exception as exc:
@@ -58,11 +61,38 @@ def validate_thumbnail(order_item_id: str) -> dict:
     return result
 
 
-def _call_azure_vision(thumbnail_path: str | None, settings) -> dict:
-    """Call Azure OpenAI GPT-4o with a base64-encoded thumbnail."""
+def _call_azure_vision(
+    thumbnail_path: str | None,
+    settings,
+    tenant_config: dict | None = None,
+) -> dict:
+    """Call Azure OpenAI GPT-4o with a base64-encoded thumbnail.
+
+    Credential resolution order:
+    1. tenant_config (if provided and ai_enabled=True)
+    2. Global settings (azure_openai_* env vars)
+    """
     import json
 
-    if not settings.azure_openai_api_key or not settings.azure_openai_endpoint:
+    # Resolve credentials from tenant config or global settings
+    if tenant_config and tenant_config.get("ai_enabled"):
+        api_key = tenant_config.get("ai_api_key") or settings.azure_openai_api_key
+        endpoint = tenant_config.get("ai_endpoint") or settings.azure_openai_endpoint
+        deployment = tenant_config.get("ai_deployment") or settings.azure_openai_deployment
+        api_version = tenant_config.get("ai_api_version") or settings.azure_openai_api_version
+        max_tokens = int(tenant_config.get("ai_max_tokens", 500))
+        temperature = float(tenant_config.get("ai_temperature", 0.1))
+        prompt = tenant_config.get("ai_validation_prompt") or VALIDATION_PROMPT
+    else:
+        api_key = settings.azure_openai_api_key
+        endpoint = settings.azure_openai_endpoint
+        deployment = settings.azure_openai_deployment
+        api_version = settings.azure_openai_api_version
+        max_tokens = 500
+        temperature = 0.1
+        prompt = VALIDATION_PROMPT
+
+    if not api_key or not endpoint:
         raise ValueError("Azure OpenAI credentials not configured")
 
     if not thumbnail_path or not Path(thumbnail_path).exists():
@@ -72,21 +102,21 @@ def _call_azure_vision(thumbnail_path: str | None, settings) -> dict:
         from openai import AzureOpenAI
 
         client = AzureOpenAI(
-            api_key=settings.azure_openai_api_key,
-            azure_endpoint=settings.azure_openai_endpoint,
-            api_version=settings.azure_openai_api_version,
+            api_key=api_key,
+            azure_endpoint=endpoint,
+            api_version=api_version,
         )
 
         with open(thumbnail_path, "rb") as f:
             image_b64 = base64.b64encode(f.read()).decode("utf-8")
 
         response = client.chat.completions.create(
-            model=settings.azure_openai_deployment,
+            model=deployment,
             messages=[
                 {
                     "role": "user",
                     "content": [
-                        {"type": "text", "text": VALIDATION_PROMPT},
+                        {"type": "text", "text": prompt},
                         {
                             "type": "image_url",
                             "image_url": {"url": f"data:image/png;base64,{image_b64}"},
@@ -94,8 +124,8 @@ def _call_azure_vision(thumbnail_path: str | None, settings) -> dict:
                     ],
                 }
             ],
-            max_tokens=500,
-            temperature=0.1,
+            max_tokens=max_tokens,
+            temperature=temperature,
         )
 
         content = response.choices[0].message.content or ""

backend/app/tasks/ai_tasks.py

@@ -7,11 +7,34 @@ logger = logging.getLogger(__name__)
 
 @celery_app.task(bind=True, name="app.tasks.ai_tasks.validate_item", queue="ai_validation")
 def validate_item(self, order_item_id: str):
-    """Validate orientation of a rendered thumbnail via Azure GPT-4o Vision."""
+    """Validate orientation of a rendered thumbnail via Azure GPT-4o Vision.
+
+    Loads the order item's tenant config and passes it to validate_thumbnail()
+    so that per-tenant Azure credentials are used when configured.
+    """
     logger.info(f"AI validation for item: {order_item_id}")
     try:
+        from app.config import settings
+        from sqlalchemy import create_engine
+        from sqlalchemy.orm import Session
+        from app.models.order_item import OrderItem
         from app.services.azure_ai import validate_thumbnail
-        validate_thumbnail(order_item_id)
+
+        # Load tenant config for this order item
+        tenant_config: dict | None = None
+        try:
+            engine = create_engine(settings.database_url_sync)
+            with Session(engine) as session:
+                item = session.get(OrderItem, __import__("uuid").UUID(order_item_id))
+                if item and hasattr(item, "order") and item.order and item.order.tenant_id:
+                    from app.domains.tenants.models import Tenant
+                    tenant = session.get(Tenant, item.order.tenant_id)
+                    if tenant:
+                        tenant_config = tenant.tenant_config or {}
+        except Exception as exc:
+            logger.warning(f"Could not load tenant config for {order_item_id}: {exc}")
+
+        validate_thumbnail(order_item_id, tenant_config=tenant_config)
     except Exception as exc:
         logger.error(f"AI validation failed for {order_item_id}: {exc}")
         raise self.retry(exc=exc, countdown=30, max_retries=3)

frontend/src/api/tenants.ts

@@ -21,6 +21,28 @@ export interface TenantUpdate {
   is_active?: boolean
 }
 
+export interface TenantAIConfig {
+  ai_enabled: boolean
+  ai_endpoint: string | null
+  ai_deployment: string
+  ai_api_version: string
+  has_api_key: boolean
+  ai_max_tokens: number
+  ai_temperature: number
+  ai_validation_prompt: string | null
+}
+
+export interface TenantAIConfigUpdate {
+  ai_enabled: boolean
+  ai_endpoint?: string | null
+  ai_deployment?: string
+  ai_api_version?: string
+  ai_api_key?: string | null
+  ai_max_tokens?: number
+  ai_temperature?: number
+  ai_validation_prompt?: string | null
+}
+
 export async function getTenants(): Promise<Tenant[]> {
   const res = await api.get<Tenant[]>('/tenants/')
   return res.data
@@ -44,3 +66,25 @@ export async function updateTenant(id: string, data: TenantUpdate): Promise<Tena
 export async function deleteTenant(id: string): Promise<void> {
   await api.delete(`/tenants/${id}`)
 }
+
+export async function getTenantAIConfig(tenantId: string): Promise<TenantAIConfig> {
+  const res = await api.get<TenantAIConfig>(`/tenants/${tenantId}/ai-config`)
+  return res.data
+}
+
+export async function updateTenantAIConfig(
+  tenantId: string,
+  config: TenantAIConfigUpdate,
+): Promise<TenantAIConfig> {
+  const res = await api.put<TenantAIConfig>(`/tenants/${tenantId}/ai-config`, config)
+  return res.data
+}
+
+export async function testTenantAIConfig(
+  tenantId: string,
+): Promise<{ ok: boolean; error?: string }> {
+  const res = await api.post<{ ok: boolean; error?: string }>(
+    `/tenants/${tenantId}/ai-config/test`,
+  )
+  return res.data
+}

frontend/src/api/worker.ts

@@ -197,3 +197,24 @@ export async function updateWorkerConfig(
   const res = await api.put<WorkerConfig>(`/worker/configs/${queueName}`, update)
   return res.data
 }
+
+// ---------------------------------------------------------------------------
+// GPU probe
+// ---------------------------------------------------------------------------
+
+export interface GPUProbeResult {
+  status: 'ok' | 'failed' | 'error' | 'unknown'
+  device_type?: string | null
+  error?: string | null
+  probed_at?: string | null
+}
+
+export async function getGpuProbeResult(): Promise<GPUProbeResult> {
+  const res = await api.get<GPUProbeResult>('/worker/gpu-probe')
+  return res.data
+}
+
+export async function triggerGpuProbe(): Promise<{ task_id: string; queued: boolean }> {
+  const res = await api.post<{ task_id: string; queued: boolean }>('/worker/gpu-probe')
+  return res.data
+}

frontend/src/pages/Admin.tsx

@@ -1,7 +1,7 @@
 import { useQuery, useMutation, useQueryClient } from '@tanstack/react-query'
-import { useState } from 'react'
+import { useState, useRef } from 'react'
 import { toast } from 'sonner'
-import { UserPlus, Trash2, Pencil, ChevronDown, ChevronUp, ChevronRight, Settings, RefreshCw, CheckCircle2, XCircle, Clock, DollarSign, Layers, AlertTriangle, Upload, FileBox, Plus, X, LayoutDashboard } from 'lucide-react'
+import { UserPlus, Trash2, Pencil, ChevronDown, ChevronUp, ChevronRight, Settings, RefreshCw, CheckCircle2, XCircle, Clock, DollarSign, Layers, AlertTriangle, Upload, FileBox, Plus, X, LayoutDashboard, Cpu, Zap } from 'lucide-react'
 import { Link } from 'react-router-dom'
 import api from '../api/client'
 import ConfirmModal from '../components/ConfirmModal'
@@ -20,6 +20,8 @@ import {
 import { getTenantDefaultDashboard } from '../api/dashboard'
 import type { WidgetConfig } from '../api/dashboard'
 import DashboardCustomizeModal from '../components/dashboard/DashboardCustomizeModal'
+import { getGpuProbeResult, triggerGpuProbe } from '../api/worker'
+import type { GPUProbeResult } from '../api/worker'
 
 export default function AdminPage() {
   const qc = useQueryClient()
@@ -202,6 +204,67 @@ export default function AdminPage() {
     staleTime: 300_000,
   })
 
+  // GPU Probe
+  const [gpuProbeExpanded, setGpuProbeExpanded] = useState(false)
+  const [gpuProbing, setGpuProbing] = useState(false)
+  const gpuPollRef = useRef<ReturnType<typeof setInterval> | null>(null)
+  const { data: gpuProbeResult, refetch: refetchGpuProbe } = useQuery<GPUProbeResult>({
+    queryKey: ['gpu-probe-result'],
+    queryFn: getGpuProbeResult,
+    enabled: isAdmin,
+    refetchInterval: gpuProbing ? 2000 : false,
+    staleTime: 0,
+  })
+
+  const handleRunGpuCheck = async () => {
+    if (!isAdmin) return
+    setGpuProbing(true)
+    try {
+      await triggerGpuProbe()
+      // Poll for up to 45 seconds
+      let elapsed = 0
+      const interval = setInterval(async () => {
+        elapsed += 2
+        await refetchGpuProbe()
+        if (elapsed >= 45) {
+          clearInterval(interval)
+          setGpuProbing(false)
+        }
+      }, 2000)
+      gpuPollRef.current = interval
+    } catch {
+      toast.error('Failed to trigger GPU check')
+      setGpuProbing(false)
+    }
+  }
+
+  const gpuStatusBadge = () => {
+    if (!gpuProbeResult) return null
+    const s = gpuProbeResult.status
+    if (s === 'ok') {
+      return (
+        <span className="inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-status-success-bg text-status-success-text">
+          <CheckCircle2 size={11} />
+          GPU OK{gpuProbeResult.device_type ? ` (${gpuProbeResult.device_type})` : ''}
+        </span>
+      )
+    }
+    if (s === 'failed') {
+      return (
+        <span className="inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-yellow-100 text-yellow-800 dark:bg-yellow-900 dark:text-yellow-200">
+          <Cpu size={11} />
+          CPU Fallback
+        </span>
+      )
+    }
+    return (
+      <span className="inline-flex items-center gap-1 px-2 py-0.5 rounded-full text-xs font-medium bg-status-error-bg text-status-error-text">
+        <XCircle size={11} />
+        {s === 'error' ? 'Error' : 'Unknown'}
+      </span>
+    )
+  }
+
   return (
     <div className="p-8 space-y-8">
       <h1 className="text-2xl font-bold text-content">Admin</h1>
@@ -1523,6 +1586,95 @@ function AssetLibraryPanel() {
         </div>
       )}
 
+      {/* ------------------------------------------------------------------ */}
+      {/* GPU Status (admin only)                                             */}
+      {/* ------------------------------------------------------------------ */}
+      {isAdmin && (
+        <div className="card">
+          <button
+            className="w-full p-4 flex items-center justify-between text-left"
+            onClick={() => setGpuProbeExpanded((v) => !v)}
+          >
+            <div className="flex items-center gap-2">
+              <Zap size={16} className="text-content-muted" />
+              <div>
+                <h2 className="font-semibold text-content">GPU Status</h2>
+                <p className="text-xs text-content-muted mt-0.5">
+                  Check Blender GPU availability on the render worker
+                </p>
+              </div>
+            </div>
+            <div className="flex items-center gap-3">
+              {gpuStatusBadge()}
+              {gpuProbeResult?.probed_at && (
+                <span className="text-xs text-content-muted">
+                  Last checked: {Math.round((Date.now() - new Date(gpuProbeResult.probed_at).getTime()) / 60000)} min ago
+                </span>
+              )}
+              {gpuProbeExpanded ? <ChevronUp size={16} className="text-content-muted" /> : <ChevronDown size={16} className="text-content-muted" />}
+            </div>
+          </button>
+
+          {gpuProbeExpanded && (
+            <div className="px-6 pb-6 space-y-4 border-t border-border-default pt-4">
+              <div className="flex items-center gap-3">
+                <button
+                  onClick={handleRunGpuCheck}
+                  disabled={gpuProbing}
+                  className="flex items-center gap-2 px-4 py-2 text-sm rounded-md bg-accent text-accent-text font-medium hover:bg-accent-hover disabled:opacity-50 disabled:cursor-not-allowed transition-colors"
+                >
+                  {gpuProbing
+                    ? <RefreshCw size={14} className="animate-spin" />
+                    : <Zap size={14} />
+                  }
+                  {gpuProbing ? 'Checking…' : 'Run GPU Check'}
+                </button>
+                {gpuProbing && (
+                  <span className="text-xs text-content-muted">
+                    Polling for result (up to 45s)…
+                  </span>
+                )}
+              </div>
+
+              {gpuProbeResult && (
+                <div className="rounded-lg border border-border-default bg-surface-alt p-4 space-y-2 text-sm">
+                  <div className="flex items-center gap-2">
+                    <span className="text-content-secondary w-28 shrink-0">Status</span>
+                    {gpuStatusBadge()}
+                  </div>
+                  {gpuProbeResult.device_type && (
+                    <div className="flex items-center gap-2">
+                      <span className="text-content-secondary w-28 shrink-0">Device type</span>
+                      <span className="text-content font-mono text-xs">{gpuProbeResult.device_type}</span>
+                    </div>
+                  )}
+                  {gpuProbeResult.error && (
+                    <div className="flex items-start gap-2">
+                      <span className="text-content-secondary w-28 shrink-0">Error</span>
+                      <span className="text-status-error-text text-xs">{gpuProbeResult.error}</span>
+                    </div>
+                  )}
+                  {gpuProbeResult.probed_at && (
+                    <div className="flex items-center gap-2">
+                      <span className="text-content-secondary w-28 shrink-0">Probed at</span>
+                      <span className="text-content-muted text-xs">
+                        {new Date(gpuProbeResult.probed_at).toLocaleString()}
+                      </span>
+                    </div>
+                  )}
+                </div>
+              )}
+
+              {!gpuProbeResult && !gpuProbing && (
+                <p className="text-sm text-content-muted">
+                  No probe result yet. Click "Run GPU Check" to trigger a check on the render worker.
+                </p>
+              )}
+            </div>
+          )}
+        </div>
+      )}
+
       <ConfirmModal
         open={confirmState.open}
         title={confirmState.title}

frontend/src/pages/Tenants.tsx

@@ -3,11 +3,13 @@ import { useQuery, useMutation, useQueryClient } from '@tanstack/react-query'
 import { toast } from 'sonner'
 import {
   Plus, Trash2, Pencil, X, Building2, ChevronDown, Check, Users,
+  Brain, CheckCircle2, XCircle, Loader2, Eye, EyeOff,
 } from 'lucide-react'
 import {
   getTenants, createTenant, updateTenant, deleteTenant,
+  getTenantAIConfig, updateTenantAIConfig, testTenantAIConfig,
 } from '../api/tenants'
-import type { Tenant, TenantCreate, TenantUpdate } from '../api/tenants'
+import type { Tenant, TenantCreate, TenantUpdate, TenantAIConfig, TenantAIConfigUpdate } from '../api/tenants'
 
 const TENANT_CONTEXT_KEY = 'schaeffler_tenant_id'
 
@@ -56,6 +58,105 @@ export default function TenantsPage() {
   // --- Delete confirm state ---
   const [deletingId, setDeletingId] = useState<string | null>(null)
 
+  // --- AI Config panel state ---
+  const [aiConfigTenant, setAIConfigTenant] = useState<Tenant | null>(null)
+  const [aiForm, setAIForm] = useState<TenantAIConfigUpdate>({
+    ai_enabled: false,
+    ai_endpoint: null,
+    ai_deployment: 'gpt-4o',
+    ai_api_version: '2024-02-01',
+    ai_api_key: null,
+    ai_max_tokens: 500,
+    ai_temperature: 0.1,
+    ai_validation_prompt: null,
+  })
+  const [showApiKey, setShowApiKey] = useState(false)
+  const [urlPasteValue, setUrlPasteValue] = useState('')
+  const [testResult, setTestResult] = useState<{ ok: boolean; error?: string } | null>(null)
+  const [isTesting, setIsTesting] = useState(false)
+
+  const { data: aiConfig } = useQuery<TenantAIConfig>({
+    queryKey: ['tenant-ai-config', aiConfigTenant?.id],
+    queryFn: () => getTenantAIConfig(aiConfigTenant!.id),
+    enabled: !!aiConfigTenant,
+  })
+
+  // Sync aiForm when aiConfig loads
+  useEffect(() => {
+    if (aiConfig) {
+      setAIForm({
+        ai_enabled: aiConfig.ai_enabled,
+        ai_endpoint: aiConfig.ai_endpoint ?? null,
+        ai_deployment: aiConfig.ai_deployment,
+        ai_api_version: aiConfig.ai_api_version,
+        ai_api_key: null,  // never pre-fill — has_api_key tells us if one is set
+        ai_max_tokens: aiConfig.ai_max_tokens,
+        ai_temperature: aiConfig.ai_temperature,
+        ai_validation_prompt: aiConfig.ai_validation_prompt ?? null,
+      })
+    }
+  }, [aiConfig])
+
+  const saveAIMut = useMutation({
+    mutationFn: ({ id, data }: { id: string; data: TenantAIConfigUpdate }) =>
+      updateTenantAIConfig(id, data),
+    onSuccess: () => {
+      toast.success('AI config saved')
+      qc.invalidateQueries({ queryKey: ['tenant-ai-config', aiConfigTenant?.id] })
+    },
+    onError: (e: any) => toast.error(e.response?.data?.detail || 'Failed to save AI config'),
+  })
+
+  const handleUrlPaste = (raw: string) => {
+    setUrlPasteValue(raw)
+    try {
+      // e.g. https://myinstance.openai.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2024-02-01
+      const url = new URL(raw)
+      const endpoint = `${url.protocol}//${url.hostname}`
+      const parts = url.pathname.split('/')
+      // /openai/deployments/{deployment}/...
+      const deployIdx = parts.indexOf('deployments')
+      const deployment = deployIdx >= 0 ? parts[deployIdx + 1] : null
+      const apiVersion = url.searchParams.get('api-version') ?? null
+      setAIForm((prev) => ({
+        ...prev,
+        ai_endpoint: endpoint || prev.ai_endpoint,
+        ...(deployment ? { ai_deployment: deployment } : {}),
+        ...(apiVersion ? { ai_api_version: apiVersion } : {}),
+      }))
+    } catch {
+      // Not a valid URL yet — ignore
+    }
+  }
+
+  const handleTestConnection = async () => {
+    if (!aiConfigTenant) return
+    setIsTesting(true)
+    setTestResult(null)
+    try {
+      const result = await testTenantAIConfig(aiConfigTenant.id)
+      setTestResult(result)
+    } catch (e: any) {
+      setTestResult({ ok: false, error: e.response?.data?.detail || 'Request failed' })
+    } finally {
+      setIsTesting(false)
+    }
+  }
+
+  const openAIConfig = (tenant: Tenant) => {
+    setAIConfigTenant(tenant)
+    setTestResult(null)
+    setShowApiKey(false)
+    setUrlPasteValue('')
+  }
+
+  const closeAIConfig = () => {
+    setAIConfigTenant(null)
+    setTestResult(null)
+    setShowApiKey(false)
+    setUrlPasteValue('')
+  }
+
   const { data: tenants = [], isLoading } = useQuery({
     queryKey: ['tenants'],
     queryFn: getTenants,
@@ -241,6 +342,13 @@ export default function TenantsPage() {
                 <td className="px-4 py-3 text-content-muted">{formatDate(tenant.created_at)}</td>
                 <td className="px-4 py-3">
                   <div className="flex items-center justify-end gap-1">
+                    <button
+                      onClick={() => openAIConfig(tenant)}
+                      className="p-1.5 rounded hover:bg-surface-alt text-content-muted hover:text-content transition-colors"
+                      title="Configure Azure AI"
+                    >
+                      <Brain size={15} />
+                    </button>
                     <button
                       onClick={() => openEdit(tenant)}
                       className="p-1.5 rounded hover:bg-surface-alt text-content-muted hover:text-content transition-colors"
@@ -404,6 +512,226 @@ export default function TenantsPage() {
         </div>
       )}
 
+      {/* AI Config Modal */}
+      {aiConfigTenant && (
+        <div className="fixed inset-0 z-50 flex items-center justify-center bg-black/40">
+          <div className="bg-surface rounded-xl shadow-xl w-full max-w-lg mx-4 p-6 max-h-[90vh] overflow-y-auto">
+            <div className="flex items-center justify-between mb-5">
+              <div className="flex items-center gap-2">
+                <Brain size={18} className="text-accent" />
+                <h2 className="text-lg font-semibold text-content">
+                  Azure AI Config — {aiConfigTenant.name}
+                </h2>
+              </div>
+              <button
+                onClick={closeAIConfig}
+                className="p-1.5 rounded hover:bg-surface-alt text-content-muted transition-colors"
+              >
+                <X size={18} />
+              </button>
+            </div>
+
+            <div className="space-y-5">
+              {/* Enable toggle */}
+              <div className="flex items-center gap-3">
+                <label className="relative inline-flex items-center cursor-pointer">
+                  <input
+                    type="checkbox"
+                    checked={aiForm.ai_enabled ?? false}
+                    onChange={(e) => setAIForm((prev) => ({ ...prev, ai_enabled: e.target.checked }))}
+                    className="sr-only peer"
+                  />
+                  <div className="w-9 h-5 bg-surface-muted rounded-full peer peer-checked:bg-accent transition-colors after:content-[''] after:absolute after:top-0.5 after:left-0.5 after:bg-white after:rounded-full after:h-4 after:w-4 after:transition-all peer-checked:after:translate-x-4" />
+                </label>
+                <span className="text-sm font-medium text-content-secondary">Azure AI Validation</span>
+              </div>
+
+              {/* URL paste helper */}
+              <div>
+                <label className="block text-xs font-semibold text-content-muted uppercase tracking-wider mb-1">
+                  Paste Azure URL (auto-fills fields below)
+                </label>
+                <input
+                  type="text"
+                  value={urlPasteValue}
+                  onChange={(e) => handleUrlPaste(e.target.value)}
+                  placeholder="https://myinstance.openai.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2024-02-01"
+                  className="w-full px-3 py-2 rounded-md border border-border-default bg-surface-alt text-content text-xs font-mono focus:outline-none focus:ring-2 focus:ring-accent/50"
+                />
+              </div>
+
+              <div className="border-t border-border-default pt-4 space-y-4">
+                {/* Endpoint */}
+                <div>
+                  <label className="block text-sm font-medium text-content-secondary mb-1">
+                    Endpoint
+                    <span className="text-xs font-normal text-content-muted ml-1">
+                      (e.g. https://myinstance.openai.azure.com)
+                    </span>
+                  </label>
+                  <input
+                    type="text"
+                    value={aiForm.ai_endpoint ?? ''}
+                    onChange={(e) => setAIForm((prev) => ({ ...prev, ai_endpoint: e.target.value || null }))}
+                    placeholder="https://myinstance.openai.azure.com"
+                    className="w-full px-3 py-2 rounded-md border border-border-default bg-surface text-content text-sm focus:outline-none focus:ring-2 focus:ring-accent/50"
+                  />
+                </div>
+
+                {/* Deployment */}
+                <div>
+                  <label className="block text-sm font-medium text-content-secondary mb-1">
+                    Deployment Name
+                  </label>
+                  <input
+                    type="text"
+                    value={aiForm.ai_deployment ?? 'gpt-4o'}
+                    onChange={(e) => setAIForm((prev) => ({ ...prev, ai_deployment: e.target.value }))}
+                    placeholder="gpt-4o"
+                    className="w-full px-3 py-2 rounded-md border border-border-default bg-surface text-content text-sm focus:outline-none focus:ring-2 focus:ring-accent/50"
+                  />
+                </div>
+
+                {/* API Version */}
+                <div>
+                  <label className="block text-sm font-medium text-content-secondary mb-1">
+                    API Version
+                  </label>
+                  <input
+                    type="text"
+                    value={aiForm.ai_api_version ?? '2024-02-01'}
+                    onChange={(e) => setAIForm((prev) => ({ ...prev, ai_api_version: e.target.value }))}
+                    placeholder="2024-02-01"
+                    className="w-full px-3 py-2 rounded-md border border-border-default bg-surface text-content text-sm font-mono focus:outline-none focus:ring-2 focus:ring-accent/50"
+                  />
+                </div>
+
+                {/* API Key */}
+                <div>
+                  <label className="block text-sm font-medium text-content-secondary mb-1">
+                    API Key
+                    {aiConfig?.has_api_key && (
+                      <span className="text-xs font-normal text-content-muted ml-2">
+                        (already set — enter new value to replace)
+                      </span>
+                    )}
+                  </label>
+                  <div className="relative">
+                    <input
+                      type={showApiKey ? 'text' : 'password'}
+                      value={aiForm.ai_api_key ?? ''}
+                      onChange={(e) => setAIForm((prev) => ({ ...prev, ai_api_key: e.target.value || null }))}
+                      placeholder={aiConfig?.has_api_key ? '●●●●●●●●●●●●' : 'Enter API key'}
+                      className="w-full px-3 py-2 pr-10 rounded-md border border-border-default bg-surface text-content text-sm focus:outline-none focus:ring-2 focus:ring-accent/50"
+                    />
+                    <button
+                      type="button"
+                      onClick={() => setShowApiKey((v) => !v)}
+                      className="absolute right-2.5 top-1/2 -translate-y-1/2 text-content-muted hover:text-content transition-colors"
+                    >
+                      {showApiKey ? <EyeOff size={15} /> : <Eye size={15} />}
+                    </button>
+                  </div>
+                </div>
+
+                {/* Advanced: max_tokens and temperature */}
+                <div className="grid grid-cols-2 gap-3">
+                  <div>
+                    <label className="block text-sm font-medium text-content-secondary mb-1">
+                      Max Tokens
+                    </label>
+                    <input
+                      type="number"
+                      min={1}
+                      max={4096}
+                      value={aiForm.ai_max_tokens ?? 500}
+                      onChange={(e) => setAIForm((prev) => ({ ...prev, ai_max_tokens: Number(e.target.value) }))}
+                      className="w-full px-3 py-2 rounded-md border border-border-default bg-surface text-content text-sm focus:outline-none focus:ring-2 focus:ring-accent/50"
+                    />
+                  </div>
+                  <div>
+                    <label className="block text-sm font-medium text-content-secondary mb-1">
+                      Temperature
+                    </label>
+                    <input
+                      type="number"
+                      min={0}
+                      max={2}
+                      step={0.1}
+                      value={aiForm.ai_temperature ?? 0.1}
+                      onChange={(e) => setAIForm((prev) => ({ ...prev, ai_temperature: Number(e.target.value) }))}
+                      className="w-full px-3 py-2 rounded-md border border-border-default bg-surface text-content text-sm focus:outline-none focus:ring-2 focus:ring-accent/50"
+                    />
+                  </div>
+                </div>
+
+                {/* Custom validation prompt */}
+                <div>
+                  <label className="block text-sm font-medium text-content-secondary mb-1">
+                    Custom Validation Prompt
+                    <span className="text-xs font-normal text-content-muted ml-1">
+                      (leave empty to use default)
+                    </span>
+                  </label>
+                  <textarea
+                    rows={4}
+                    value={aiForm.ai_validation_prompt ?? ''}
+                    onChange={(e) => setAIForm((prev) => ({ ...prev, ai_validation_prompt: e.target.value || null }))}
+                    placeholder="Optional: override the default Schaeffler bearing analysis prompt"
+                    className="w-full px-3 py-2 rounded-md border border-border-default bg-surface text-content text-sm resize-y focus:outline-none focus:ring-2 focus:ring-accent/50"
+                  />
+                </div>
+              </div>
+
+              {/* Test connection result */}
+              {testResult && (
+                <div className={`flex items-center gap-2 px-3 py-2 rounded-md text-sm ${
+                  testResult.ok
+                    ? 'bg-status-success-bg text-status-success-text'
+                    : 'bg-status-error-bg text-status-error-text'
+                }`}>
+                  {testResult.ok
+                    ? <CheckCircle2 size={15} />
+                    : <XCircle size={15} />
+                  }
+                  <span>{testResult.ok ? 'Connected successfully' : (testResult.error ?? 'Connection failed')}</span>
+                </div>
+              )}
+            </div>
+
+            <div className="flex items-center justify-between mt-6 pt-4 border-t border-border-default">
+              <button
+                onClick={handleTestConnection}
+                disabled={isTesting || !aiConfig?.has_api_key && !aiForm.ai_api_key}
+                className="flex items-center gap-2 px-4 py-2 text-sm rounded-md border border-border-default text-content-secondary hover:bg-surface-alt disabled:opacity-50 disabled:cursor-not-allowed transition-colors"
+              >
+                {isTesting
+                  ? <Loader2 size={15} className="animate-spin" />
+                  : <CheckCircle2 size={15} />
+                }
+                Test Connection
+              </button>
+
+              <div className="flex gap-2">
+                <button
+                  onClick={closeAIConfig}
+                  className="px-4 py-2 text-sm rounded-md border border-border-default text-content-secondary hover:bg-surface-alt transition-colors"
+                >
+                  Cancel
+                </button>
+                <button
+                  onClick={() => saveAIMut.mutate({ id: aiConfigTenant.id, data: aiForm })}
+                  disabled={saveAIMut.isPending}
+                  className="px-4 py-2 text-sm rounded-md bg-accent text-accent-text font-medium hover:bg-accent-hover disabled:opacity-50 disabled:cursor-not-allowed transition-colors"
+                >
+                  {saveAIMut.isPending ? 'Saving…' : 'Save'}
+                </button>
+              </div>
+            </div>
+          </div>
+        </div>
+      )}
+
       {/* Delete Confirm Modal */}
       {deletingId && (
         <div className="fixed inset-0 z-50 flex items-center justify-center bg-black/40">