wip: AI chat integration with multi-provider support

- Add AI service with Anthropic, OpenAI, and Ollama providers - Add AI chat UI component with streaming responses - Add AI settings page for configuration - Add agent exec framework for command execution - Add API endpoints for AI chat and configuration
2026-02-18 00:17:39 +01:00 · 2025-12-04 20:16:53 +00:00
parent 59b713d176
commit 53d7776d6b
40 changed files with 7013 additions and 50 deletions
--- a/cmd/pulse-docker-agent/main.go
+++ b/cmd/pulse-docker-agent/main.go
@@ -55,6 +55,12 @@ func main() {
 	logger := zerolog.New(os.Stdout).Level(cfg.LogLevel).With().Timestamp().Logger()
 	cfg.Logger = &logger

+	// Deprecation warning
+	logger.Warn().Msg("pulse-docker-agent is DEPRECATED and will be removed in a future release")
+	logger.Warn().Msg("Please migrate to the unified 'pulse-agent' with --enable-docker flag")
+	logger.Warn().Msg("Example: pulse-agent --url <URL> --token <TOKEN> --enable-docker")
+	logger.Warn().Msg("")
+
 	agent, err := dockeragent.New(cfg)
 	if err != nil {
 		logger.Fatal().Err(err).Msg("Failed to create docker agent")
--- a/cmd/pulse-host-agent/main.go
+++ b/cmd/pulse-host-agent/main.go
@@ -61,6 +61,12 @@ func main() {

 	g, ctx := errgroup.WithContext(ctx)

+	// Deprecation warning
+	logger.Warn().Msg("pulse-host-agent is DEPRECATED and will be removed in a future release")
+	logger.Warn().Msg("Please migrate to the unified 'pulse-agent' with --enable-host flag")
+	logger.Warn().Msg("Example: pulse-agent --url <URL> --token <TOKEN> --enable-host")
+	logger.Warn().Msg("")
+
 	logger.Info().
 		Str("version", Version).
 		Str("pulse_url", hostCfg.PulseURL).
--- a/docs/API.md
+++ b/docs/API.md
@@ -112,23 +112,6 @@ Triggers a test alert to all configured channels.

 ---

---
-## 🖥️ Host Agent
-
-### Submit Report
-`POST /api/agents/host/report`
-Used by the Pulse Host Agent to push system metrics.
-
-### Lookup Agent
-`POST /api/agents/host/lookup`
-Check if a host agent is already registered.
-
-### Delete Host
-`DELETE /api/agents/host/<id>`
-Remove a host agent from monitoring.
-
---
-
 ## ⚙️ System Settings

 ### Get Settings
@@ -157,15 +140,23 @@ Initiate OIDC login flow.

 ---

-## 🐳 Docker Agent
+## 🤖 Agent Endpoints

-### Submit Report
-`POST /api/agents/docker/report`
-Used by the Pulse Docker Agent to push container metrics.
+### Unified Agent (Recommended)
+`GET /download/pulse-agent`
+Downloads the unified agent binary for the current platform.

-### Download Agent
-`GET /download/pulse-docker-agent`
-Downloads the binary for the current platform.
+The unified agent combines host and Docker monitoring. Use `--enable-docker` to enable Docker metrics.
+
+See [UNIFIED_AGENT.md](UNIFIED_AGENT.md) for installation instructions.
+
+### Legacy Agents (Deprecated)
+`GET /download/pulse-host-agent` - *Deprecated, use pulse-agent*
+`GET /download/pulse-docker-agent` - *Deprecated, use pulse-agent --enable-docker*
+
+### Submit Reports
+`POST /api/agents/host/report` - Host metrics
+`POST /api/agents/docker/report` - Docker container metrics

 ---

--- a/frontend-modern/package-lock.json
+++ b/frontend-modern/package-lock.json
@@ -11,10 +11,12 @@
      "dependencies": {
        "@solidjs/router": "^0.10.10",
        "lucide-solid": "^0.545.0",
+        "marked": "^17.0.1",
        "solid-js": "^1.8.0"
      },
      "devDependencies": {
        "@solidjs/testing-library": "^0.8.5",
+        "@tailwindcss/typography": "^0.5.19",
        "@testing-library/jest-dom": "^6.5.0",
        "@types/node": "^20.10.0",
        "@typescript-eslint/eslint-plugin": "^8.0.0",
@@ -1429,6 +1431,33 @@
        }
      }
    },
+    "node_modules/@tailwindcss/typography": {
+      "version": "0.5.19",
+      "resolved": "https://registry.npmjs.org/@tailwindcss/typography/-/typography-0.5.19.tgz",
+      "integrity": "sha512-w31dd8HOx3k9vPtcQh5QHP9GwKcgbMp87j58qi6xgiBnFFtKEAgCWnDw4qUT8aHwkCp8bKvb/KGKWWHedP0AAg==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "postcss-selector-parser": "6.0.10"
+      },
+      "peerDependencies": {
+        "tailwindcss": ">=3.0.0 || insiders || >=4.0.0-alpha.20 || >=4.0.0-beta.1"
+      }
+    },
+    "node_modules/@tailwindcss/typography/node_modules/postcss-selector-parser": {
+      "version": "6.0.10",
+      "resolved": "https://registry.npmjs.org/postcss-selector-parser/-/postcss-selector-parser-6.0.10.tgz",
+      "integrity": "sha512-IQ7TZdoaqbT+LCpShg46jnZVlhWD2w6iQYAcYXfHARZ7X1t/UGhhceQDs5X0cGqKvYlHNOuv7Oa1xmb0oQuA3w==",
+      "dev": true,
+      "license": "MIT",
+      "dependencies": {
+        "cssesc": "^3.0.0",
+        "util-deprecate": "^1.0.2"
+      },
+      "engines": {
+        "node": ">=4"
+      }
+    },
    "node_modules/@testing-library/dom": {
      "version": "10.4.1",
      "resolved": "https://registry.npmjs.org/@testing-library/dom/-/dom-10.4.1.tgz",
@@ -3895,6 +3924,18 @@
        "@jridgewell/sourcemap-codec": "^1.5.5"
      }
    },
+    "node_modules/marked": {
+      "version": "17.0.1",
+      "resolved": "https://registry.npmjs.org/marked/-/marked-17.0.1.tgz",
+      "integrity": "sha512-boeBdiS0ghpWcSwoNm/jJBwdpFaMnZWRzjA6SkUMYb40SVaN1x7mmfGKp0jvexGcx+7y2La5zRZsYFZI6Qpypg==",
+      "license": "MIT",
+      "bin": {
+        "marked": "bin/marked.js"
+      },
+      "engines": {
+        "node": ">= 20"
+      }
+    },
    "node_modules/math-intrinsics": {
      "version": "1.1.0",
      "resolved": "https://registry.npmjs.org/math-intrinsics/-/math-intrinsics-1.1.0.tgz",
@@ -5113,6 +5154,7 @@
      "integrity": "sha512-6A2rnmW5xZMdw11LYjhcI5846rt9pbLSabY5XPxo+XWdxwZaFEn47Go4NzFiHu9sNNmr/kXivP1vStfvMaK1GQ==",
      "dev": true,
      "license": "MIT",
+      "peer": true,
      "dependencies": {
        "@alloc/quick-lru": "^5.2.0",
        "arg": "^5.0.2",
--- a/frontend-modern/package.json
+++ b/frontend-modern/package.json
@@ -28,10 +28,12 @@
  "dependencies": {
    "@solidjs/router": "^0.10.10",
    "lucide-solid": "^0.545.0",
+    "marked": "^17.0.1",
    "solid-js": "^1.8.0"
  },
  "devDependencies": {
    "@solidjs/testing-library": "^0.8.5",
+    "@tailwindcss/typography": "^0.5.19",
    "@testing-library/jest-dom": "^6.5.0",
    "@types/node": "^20.10.0",
    "@typescript-eslint/eslint-plugin": "^8.0.0",
--- a/frontend-modern/src/App.tsx
+++ b/frontend-modern/src/App.tsx
@@ -43,6 +43,8 @@ import { TokenRevealDialog } from './components/TokenRevealDialog';
 import { useAlertsActivation } from './stores/alertsActivation';
 import { UpdateProgressModal } from './components/UpdateProgressModal';
 import type { UpdateStatus } from './api/updates';
+import { AIChat } from './components/AI/AIChat';
+import { aiChatStore } from './stores/aiChat';

 const Dashboard = lazy(() =>
  import('./components/Dashboard/Dashboard').then((module) => ({ default: module.Dashboard })),
@@ -723,6 +725,39 @@ function App() {

  // Root layout component for Router
  const RootLayout = (props: { children?: JSX.Element }) => {
+    // Check AI settings on mount and setup keyboard shortcut
+    onMount(() => {
+      // Check if AI is enabled
+      import('./api/ai').then(({ AIAPI }) => {
+        AIAPI.getSettings()
+          .then((settings) => {
+            aiChatStore.setEnabled(settings.enabled && settings.configured);
+          })
+          .catch(() => {
+            aiChatStore.setEnabled(false);
+          });
+      });
+
+      // Keyboard shortcut: Cmd/Ctrl+K to toggle AI
+      const handleKeyDown = (e: KeyboardEvent) => {
+        if ((e.metaKey || e.ctrlKey) && e.key === 'k') {
+          e.preventDefault();
+          if (aiChatStore.enabled) {
+            aiChatStore.toggle();
+          }
+        }
+        // Escape to close
+        if (e.key === 'Escape' && aiChatStore.isOpen) {
+          aiChatStore.close();
+        }
+      };
+
+      document.addEventListener('keydown', handleKeyDown);
+      onCleanup(() => {
+        document.removeEventListener('keydown', handleKeyDown);
+      });
+    });
+
    return (
      <Show
        when={!isLoading()}
@@ -741,24 +776,62 @@ function App() {
                  <DemoBanner />
                  <UpdateBanner />
                  <GlobalUpdateProgressWatcher />
-                  <div class="min-h-screen bg-gray-100 dark:bg-gray-900 text-gray-800 dark:text-gray-200 font-sans py-4 sm:py-6">
-                    <AppLayout
-                      connected={connected}
-                      reconnecting={reconnecting}
-                      dataUpdated={dataUpdated}
-                      lastUpdateText={lastUpdateText}
-                      versionInfo={versionInfo}
-                      hasAuth={hasAuth}
-                      needsAuth={needsAuth}
-                      proxyAuthInfo={proxyAuthInfo}
-                      handleLogout={handleLogout}
-                      state={state}
-                    >
-                      {props.children}
-                    </AppLayout>
+                  {/* Main layout container - flexbox to allow AI panel to push content */}
+                  <div class="flex h-screen overflow-hidden">
+                    {/* Main content area - shrinks when AI panel is open, scrolls independently */}
+                    <div class={`flex-1 min-w-0 overflow-y-auto bg-gray-100 dark:bg-gray-900 text-gray-800 dark:text-gray-200 font-sans py-4 sm:py-6 transition-all duration-300`}>
+                      <AppLayout
+                        connected={connected}
+                        reconnecting={reconnecting}
+                        dataUpdated={dataUpdated}
+                        lastUpdateText={lastUpdateText}
+                        versionInfo={versionInfo}
+                        hasAuth={hasAuth}
+                        needsAuth={needsAuth}
+                        proxyAuthInfo={proxyAuthInfo}
+                        handleLogout={handleLogout}
+                        state={state}
+                      >
+                        {props.children}
+                      </AppLayout>
+                    </div>
+                    {/* AI Panel - slides in from right, pushes content */}
+                    <AIChat onClose={() => aiChatStore.close()} />
                  </div>
                  <ToastContainer />
                  <TokenRevealDialog />
+                  {/* Fixed AI Assistant Button - always visible on the side when AI is enabled */}
+                  <Show when={aiChatStore.enabled !== false && !aiChatStore.isOpen}>
+                    {/* This component only shows when chat is closed */}
+                    <button
+                      type="button"
+                      onClick={() => aiChatStore.toggle()}
+                      class="fixed right-0 top-1/2 -translate-y-1/2 z-40 flex items-center gap-1.5 pl-2 pr-1.5 py-3 rounded-l-xl bg-gradient-to-r from-purple-600 to-purple-700 text-white shadow-lg hover:from-purple-700 hover:to-purple-800 transition-all duration-200 group"
+                      title={aiChatStore.context.context?.name ? `AI Assistant - ${aiChatStore.context.context.name}` : 'AI Assistant (⌘K)'}
+                      aria-label="Expand AI Assistant"
+                    >
+                      {/* Double chevron left - expand */}
+                      <svg
+                        class="h-4 w-4 flex-shrink-0"
+                        fill="none"
+                        stroke="currentColor"
+                        viewBox="0 0 24 24"
+                        stroke-width="2"
+                      >
+                        <path
+                          stroke-linecap="round"
+                          stroke-linejoin="round"
+                          d="M11 19l-7-7 7-7M18 19l-7-7 7-7"
+                        />
+                      </svg>
+                      {/* Context indicator - shows count when items are in context */}
+                      <Show when={aiChatStore.contextItems.length > 0}>
+                        <span class="min-w-[18px] h-[18px] px-1 flex items-center justify-center text-[10px] font-bold bg-green-500 text-white rounded-full">
+                          {aiChatStore.contextItems.length}
+                        </span>
+                      </Show>
+                    </button>
+                  </Show>
                  <TooltipRoot />
                </DarkModeContext.Provider>
              </WebSocketContext.Provider>
--- a/frontend-modern/src/api/ai.ts
+++ b/frontend-modern/src/api/ai.ts
@@ -0,0 +1,127 @@
+import { apiFetchJSON, apiFetch } from '@/utils/apiClient';
+import type {
+  AISettings,
+  AISettingsUpdateRequest,
+  AITestResult,
+  AIExecuteRequest,
+  AIExecuteResponse,
+  AIStreamEvent,
+} from '@/types/ai';
+
+export class AIAPI {
+  private static baseUrl = '/api';
+
+  // Get AI settings
+  static async getSettings(): Promise<AISettings> {
+    return apiFetchJSON(`${this.baseUrl}/settings/ai`) as Promise<AISettings>;
+  }
+
+  // Update AI settings
+  static async updateSettings(settings: AISettingsUpdateRequest): Promise<AISettings> {
+    return apiFetchJSON(`${this.baseUrl}/settings/ai/update`, {
+      method: 'PUT',
+      body: JSON.stringify(settings),
+    }) as Promise<AISettings>;
+  }
+
+  // Test AI connection
+  static async testConnection(): Promise<AITestResult> {
+    return apiFetchJSON(`${this.baseUrl}/ai/test`, {
+      method: 'POST',
+    }) as Promise<AITestResult>;
+  }
+
+  // Execute an AI prompt
+  static async execute(request: AIExecuteRequest): Promise<AIExecuteResponse> {
+    return apiFetchJSON(`${this.baseUrl}/ai/execute`, {
+      method: 'POST',
+      body: JSON.stringify(request),
+    }) as Promise<AIExecuteResponse>;
+  }
+
+  // Run a single command (for approved commands)
+  static async runCommand(request: {
+    command: string;
+    target_type: string;
+    target_id: string;
+    run_on_host: boolean;
+    vmid?: string;
+  }): Promise<{ output: string; success: boolean; error?: string }> {
+    return apiFetchJSON(`${this.baseUrl}/ai/run-command`, {
+      method: 'POST',
+      body: JSON.stringify(request),
+    }) as Promise<{ output: string; success: boolean; error?: string }>;
+  }
+
+  // Execute an AI prompt with streaming
+  // Returns an abort function to cancel the request
+  static async executeStream(
+    request: AIExecuteRequest,
+    onEvent: (event: AIStreamEvent) => void,
+    signal?: AbortSignal
+  ): Promise<void> {
+    console.log('[AI SSE] Starting streaming request:', request);
+
+    const response = await apiFetch(`${this.baseUrl}/ai/execute/stream`, {
+      method: 'POST',
+      body: JSON.stringify(request),
+      headers: {
+        'Content-Type': 'application/json',
+        'Accept': 'text/event-stream',
+      },
+      signal,
+    });
+
+    console.log('[AI SSE] Response status:', response.status, response.headers.get('content-type'));
+
+    if (!response.ok) {
+      const text = await response.text();
+      console.error('[AI SSE] Request failed:', text);
+      throw new Error(text || `Request failed with status ${response.status}`);
+    }
+
+    const reader = response.body?.getReader();
+    if (!reader) {
+      console.error('[AI SSE] No response body');
+      throw new Error('No response body');
+    }
+
+    const decoder = new TextDecoder();
+    let buffer = '';
+
+    console.log('[AI SSE] Starting to read stream...');
+
+    try {
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) {
+          console.log('[AI SSE] Stream ended');
+          break;
+        }
+
+        const chunk = decoder.decode(value, { stream: true });
+        console.log('[AI SSE] Received chunk:', chunk.length, 'bytes');
+        buffer += chunk;
+
+        // Process complete SSE messages
+        const lines = buffer.split('\n\n');
+        buffer = lines.pop() || ''; // Keep incomplete message in buffer
+
+        for (const line of lines) {
+          if (line.startsWith('data: ')) {
+            try {
+              const data = JSON.parse(line.slice(6));
+              console.log('[AI SSE] Parsed event:', data.type, data);
+              onEvent(data as AIStreamEvent);
+            } catch (e) {
+              console.error('[AI SSE] Failed to parse event:', e, line);
+            }
+          }
+        }
+      }
+    } finally {
+      reader.releaseLock();
+      console.log('[AI SSE] Reader released');
+    }
+  }
+}
--- a/frontend-modern/src/api/dockerMetadata.ts
+++ b/frontend-modern/src/api/dockerMetadata.ts
@@ -6,6 +6,7 @@ export interface DockerMetadata {
  customUrl?: string;
  description?: string;
  tags?: string[];
+  notes?: string[]; // User annotations for AI context
 }

 export class DockerMetadataAPI {
--- a/frontend-modern/src/api/guestMetadata.ts
+++ b/frontend-modern/src/api/guestMetadata.ts
@@ -6,6 +6,7 @@ export interface GuestMetadata {
  customUrl?: string;
  description?: string;
  tags?: string[];
+  notes?: string[]; // User annotations for AI context
 }

 export class GuestMetadataAPI {
--- a/frontend-modern/src/components/AI/AIChat.tsx
+++ b/frontend-modern/src/components/AI/AIChat.tsx
--- a/frontend-modern/src/components/Dashboard/GuestDrawer.tsx
+++ b/frontend-modern/src/components/Dashboard/GuestDrawer.tsx
@@ -1,7 +1,11 @@
-import { Component, Show, For } from 'solid-js';
+import { Component, Show, For, createSignal, createEffect } from 'solid-js';
 import { VM, Container } from '@/types/api';
-import { formatBytes } from '@/utils/format';
+import { formatBytes, formatPercent, formatSpeed, formatUptime } from '@/utils/format';
 import { DiskList } from './DiskList';
+import { aiChatStore } from '@/stores/aiChat';
+import { AIAPI } from '@/api/ai';
+import { GuestMetadataAPI } from '@/api/guestMetadata';
+import { logger } from '@/utils/logger';

 type Guest = VM | Container;

@@ -12,10 +16,185 @@ interface GuestDrawerProps {
 }

 export const GuestDrawer: Component<GuestDrawerProps> = (props) => {
+    const [aiEnabled, setAiEnabled] = createSignal(false);
+    const [annotations, setAnnotations] = createSignal<string[]>([]);
+    const [newAnnotation, setNewAnnotation] = createSignal('');
+    const [saving, setSaving] = createSignal(false);
+
+    // Build guest ID for metadata
+    const guestId = () => props.guest.id || `${props.guest.instance}-${props.guest.vmid}`;
+
+    // Check if AI is enabled and load annotations on mount
+    createEffect(() => {
+        AIAPI.getSettings()
+            .then((settings) => setAiEnabled(settings.enabled && settings.configured))
+            .catch((err) => logger.debug('[GuestDrawer] AI settings check failed:', err));
+
+        // Load existing annotations
+        GuestMetadataAPI.getMetadata(guestId())
+            .then((meta) => {
+                if (meta.notes && Array.isArray(meta.notes)) setAnnotations(meta.notes);
+            })
+            .catch((err) => logger.debug('[GuestDrawer] Failed to load annotations:', err));
+    });
+
+    // Update AI context whenever guest changes or annotations are loaded
+    createEffect(() => {
+        const guestType = props.guest.type === 'qemu' ? 'vm' : 'container';
+        // Track annotations to re-run when they change
+        void annotations();
+        aiChatStore.setTargetContext(guestType, guestId(), {
+            guestName: props.guest.name,
+            ...buildGuestContext(),
+        });
+    });
+
+    // Note: We no longer clear context on unmount - context accumulates across navigation
+    // Users can clear individual items or all context from the AI panel
+
+    const saveAnnotations = async (updated: string[]) => {
+        setSaving(true);
+        try {
+            await GuestMetadataAPI.updateMetadata(guestId(), { notes: updated });
+            logger.debug('[GuestDrawer] Annotations saved');
+        } catch (err) {
+            logger.error('[GuestDrawer] Failed to save annotations:', err);
+        } finally {
+            setSaving(false);
+        }
+    };
+
+    const addAnnotation = () => {
+        const text = newAnnotation().trim();
+        if (!text) return;
+        const updated = [...annotations(), text];
+        setAnnotations(updated);
+        setNewAnnotation('');
+        saveAnnotations(updated);
+    };
+
+    const removeAnnotation = (index: number) => {
+        const updated = annotations().filter((_, i) => i !== index);
+        setAnnotations(updated);
+        saveAnnotations(updated);
+    };
+
+    const handleKeyDown = (e: KeyboardEvent) => {
+        if (e.key === 'Enter') {
+            e.preventDefault();
+            addAnnotation();
+        }
+    };
+
    const isVM = (guest: Guest): guest is VM => {
        return guest.type === 'qemu';
    };

+    // Build comprehensive context for AI
+    const buildGuestContext = () => {
+        const guest = props.guest;
+        const context: Record<string, unknown> = {
+            name: guest.name,
+            type: guest.type === 'qemu' ? 'Virtual Machine' : 'LXC Container',
+            vmid: guest.vmid,
+            node: guest.node,  // PVE node this guest lives on
+            guest_node: guest.node,  // For backend agent routing
+            status: guest.status,
+            uptime: guest.uptime ? formatUptime(guest.uptime) : 'Not running',
+        };
+
+        // CPU info
+        if (guest.cpu !== undefined) {
+            context.cpu_usage = formatPercent(guest.cpu * 100);
+        }
+        if (guest.cpus) {
+            context.cpu_cores = guest.cpus;
+        }
+
+        // Memory info
+        if (guest.memory) {
+            context.memory_used = formatBytes(guest.memory.used || 0);
+            context.memory_total = formatBytes(guest.memory.total || 0);
+            context.memory_usage = formatPercent(guest.memory.usage || 0);
+            if (guest.memory.balloon && guest.memory.balloon !== guest.memory.total) {
+                context.memory_balloon = formatBytes(guest.memory.balloon);
+            }
+            if (guest.memory.swapTotal && guest.memory.swapTotal > 0) {
+                context.swap_used = formatBytes(guest.memory.swapUsed || 0);
+                context.swap_total = formatBytes(guest.memory.swapTotal);
+            }
+        }
+
+        // Disk info
+        if (guest.disk && guest.disk.total > 0) {
+            context.disk_used = formatBytes(guest.disk.used || 0);
+            context.disk_total = formatBytes(guest.disk.total || 0);
+            context.disk_usage = formatPercent((guest.disk.used / guest.disk.total) * 100);
+        }
+
+        // I/O rates
+        if (guest.diskRead !== undefined) {
+            context.disk_read_rate = formatSpeed(guest.diskRead);
+        }
+        if (guest.diskWrite !== undefined) {
+            context.disk_write_rate = formatSpeed(guest.diskWrite);
+        }
+        if (guest.networkIn !== undefined) {
+            context.network_in_rate = formatSpeed(guest.networkIn);
+        }
+        if (guest.networkOut !== undefined) {
+            context.network_out_rate = formatSpeed(guest.networkOut);
+        }
+
+        // OS info (VMs only)
+        if (isVM(guest)) {
+            if (guest.osName) context.os_name = guest.osName;
+            if (guest.osVersion) context.os_version = guest.osVersion;
+            if (guest.agentVersion) context.guest_agent = guest.agentVersion;
+            if (guest.ipAddresses?.length) context.ip_addresses = guest.ipAddresses;
+        }
+
+        // Tags
+        if (guest.tags?.length) {
+            context.tags = guest.tags;
+        }
+
+        // Backup info - Pulse already has this from PBS
+        if (guest.lastBackup) {
+            const backupDate = new Date(guest.lastBackup);
+            const now = new Date();
+            const daysSinceBackup = Math.floor((now.getTime() - backupDate.getTime()) / (1000 * 60 * 60 * 24));
+            context.last_backup = backupDate.toISOString();
+            context.days_since_backup = daysSinceBackup;
+            if (daysSinceBackup > 30) {
+                context.backup_status = 'CRITICAL - No backup in over 30 days';
+            } else if (daysSinceBackup > 7) {
+                context.backup_status = 'Warning - Backup is over a week old';
+            } else {
+                context.backup_status = 'OK';
+            }
+        } else {
+            context.backup_status = 'NEVER - No backup recorded';
+        }
+
+        // User annotations (for AI context)
+        if (annotations().length > 0) {
+            context.user_annotations = annotations();
+        }
+
+        return context;
+    };
+
+    const handleAskAI = () => {
+        const guestType = props.guest.type === 'qemu' ? 'vm' : 'container';
+        const guestId = props.guest.id || `${props.guest.instance}-${props.guest.vmid}`;
+
+        aiChatStore.openForTarget(guestType, guestId, {
+            guestName: props.guest.name,
+            ...buildGuestContext(),
+        });
+    };
+
    const hasOsInfo = () => {
        if (!isVM(props.guest)) return false;
        return (props.guest.osName?.length ?? 0) > 0 || (props.guest.osVersion?.length ?? 0) > 0;
@@ -79,7 +258,9 @@ export const GuestDrawer: Component<GuestDrawerProps> = (props) => {
    };

    return (
-        <div class="flex items-start gap-4">
+        <div class="space-y-3">
+            {/* Top row: metrics columns */}
+            <div class="flex items-start gap-4">
            {/* Left Column: Guest Overview */}
            <div class="min-w-[220px] flex-1 rounded border border-gray-200 bg-white/70 p-2 shadow-sm dark:border-gray-600/70 dark:bg-gray-900/30">
                <div class="text-[11px] font-medium text-gray-700 dark:text-gray-200">Guest Overview</div>
@@ -194,6 +375,73 @@ export const GuestDrawer: Component<GuestDrawerProps> = (props) => {
                    </div>
                </div>
            </Show>
+            </div>
+
+            {/* Bottom row: AI Context & Ask AI - only show when AI is enabled */}
+            <Show when={aiEnabled()}>
+                <div class="mt-3 pt-3 border-t border-gray-200 dark:border-gray-700 space-y-2">
+                    <div class="flex items-center gap-1.5">
+                        <span class="text-[10px] font-medium text-gray-500 dark:text-gray-400">AI Context</span>
+                        <Show when={saving()}>
+                            <span class="text-[9px] text-gray-400">saving...</span>
+                        </Show>
+                    </div>
+
+                    {/* Existing annotations */}
+                    <Show when={annotations().length > 0}>
+                        <div class="flex flex-wrap gap-1.5">
+                            <For each={annotations()}>
+                                {(annotation, index) => (
+                                    <span class="inline-flex items-center gap-1 px-2 py-1 text-[11px] rounded-md bg-purple-100 text-purple-800 dark:bg-purple-900/40 dark:text-purple-200">
+                                        <span class="max-w-[300px] truncate">{annotation}</span>
+                                        <button
+                                            type="button"
+                                            onClick={() => removeAnnotation(index())}
+                                            class="ml-0.5 p-0.5 rounded hover:bg-purple-200 dark:hover:bg-purple-800 transition-colors"
+                                            title="Remove"
+                                        >
+                                            <svg class="w-3 h-3" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                                                <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M6 18L18 6M6 6l12 12" />
+                                            </svg>
+                                        </button>
+                                    </span>
+                                )}
+                            </For>
+                        </div>
+                    </Show>
+
+                    {/* Add new annotation + Ask AI */}
+                    <div class="flex items-center gap-2">
+                        <input
+                            type="text"
+                            value={newAnnotation()}
+                            onInput={(e) => setNewAnnotation(e.currentTarget.value)}
+                            onKeyDown={handleKeyDown}
+                            placeholder="Add context for AI (press Enter)..."
+                            class="flex-1 px-2 py-1.5 text-[11px] rounded border border-gray-200 dark:border-gray-600 bg-white dark:bg-gray-800 text-gray-700 dark:text-gray-200 placeholder-gray-400 dark:placeholder-gray-500 focus:outline-none focus:ring-1 focus:ring-purple-500 focus:border-purple-500"
+                        />
+                        <button
+                            type="button"
+                            onClick={addAnnotation}
+                            disabled={!newAnnotation().trim()}
+                            class="px-2 py-1.5 text-[11px] rounded border border-gray-200 dark:border-gray-600 bg-white dark:bg-gray-800 text-gray-700 dark:text-gray-200 hover:bg-gray-50 dark:hover:bg-gray-700 disabled:opacity-50 disabled:cursor-not-allowed transition-colors"
+                        >
+                            Add
+                        </button>
+                        <button
+                            type="button"
+                            onClick={handleAskAI}
+                            class="flex items-center gap-1.5 px-3 py-1.5 rounded-lg bg-gradient-to-r from-purple-500 to-pink-500 text-white text-[11px] font-medium shadow-sm hover:from-purple-600 hover:to-pink-600 transition-all"
+                            title={`Ask AI about ${props.guest.name}`}
+                        >
+                            <svg class="w-3.5 h-3.5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                                <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9.75 3.104v5.714a2.25 2.25 0 01-.659 1.591L5 14.5M9.75 3.104c-.251.023-.501.05-.75.082m.75-.082a24.301 24.301 0 014.5 0m0 0v5.714c0 .597.237 1.17.659 1.591L19.8 15.3M14.25 3.104c.251.023.501.05.75.082M19.8 15.3l-1.57.393A9.065 9.065 0 0112 15a9.065 9.065 0 00-6.23.693L5 14.5m14.8.8l1.402 1.402c1.232 1.232.65 3.318-1.067 3.611l-2.576.43a18.003 18.003 0 01-5.118 0l-2.576-.43c-1.717-.293-2.299-2.379-1.067-3.611L5 14.5" />
+                            </svg>
+                            Ask AI
+                        </button>
+                    </div>
+                </div>
+            </Show>
        </div>
    );
 };
--- a/frontend-modern/src/components/Docker/DockerUnifiedTable.tsx
+++ b/frontend-modern/src/components/Docker/DockerUnifiedTable.tsx
@@ -9,6 +9,8 @@ import { DockerMetadataAPI } from '@/api/dockerMetadata';
 import { resolveHostRuntime } from './runtimeDisplay';
 import { showSuccess, showError } from '@/utils/toast';
 import { logger } from '@/utils/logger';
+import { aiChatStore } from '@/stores/aiChat';
+import { AIAPI } from '@/api/ai';
 import { buildMetricKey } from '@/utils/metricsKeys';
 import { StatusDot } from '@/components/shared/StatusDot';
 import {
@@ -790,6 +792,85 @@ const DockerContainerRow: Component<{
  });
  let urlInputRef: HTMLInputElement | undefined;

+  // Annotations and AI state
+  const [aiEnabled, setAiEnabled] = createSignal(false);
+  const [annotations, setAnnotations] = createSignal<string[]>([]);
+  const [newAnnotation, setNewAnnotation] = createSignal('');
+  const [saving, setSaving] = createSignal(false);
+
+  // Check if AI is enabled and load annotations on mount
+  createEffect(() => {
+    AIAPI.getSettings()
+      .then((settings) => setAiEnabled(settings.enabled && settings.configured))
+      .catch((err) => logger.debug('[DockerContainer] AI settings check failed:', err));
+
+    // Load existing annotations
+    DockerMetadataAPI.getMetadata(resourceId())
+      .then((meta) => {
+        if (meta.notes && Array.isArray(meta.notes)) setAnnotations(meta.notes);
+      })
+      .catch((err) => logger.debug('[DockerContainer] Failed to load annotations:', err));
+  });
+
+  const saveAnnotations = async (newAnnotations: string[]) => {
+    setSaving(true);
+    try {
+      await DockerMetadataAPI.updateMetadata(resourceId(), { notes: newAnnotations });
+      logger.debug('[DockerContainer] Annotations saved');
+    } catch (err) {
+      logger.error('[DockerContainer] Failed to save annotations:', err);
+    } finally {
+      setSaving(false);
+    }
+  };
+
+  const addAnnotation = () => {
+    const text = newAnnotation().trim();
+    if (!text) return;
+    const updated = [...annotations(), text];
+    setAnnotations(updated);
+    setNewAnnotation('');
+    saveAnnotations(updated);
+  };
+
+  const removeAnnotation = (index: number) => {
+    const updated = annotations().filter((_, i) => i !== index);
+    setAnnotations(updated);
+    saveAnnotations(updated);
+  };
+
+  const handleKeyDown = (e: KeyboardEvent) => {
+    if (e.key === 'Enter' && !e.shiftKey) {
+      e.preventDefault();
+      addAnnotation();
+    }
+  };
+
+  const buildContainerContext = () => {
+    const ctx: Record<string, unknown> = {
+      name: container.name,
+      type: 'Docker Container',
+      host: host.hostname,
+      status: container.status || container.state,
+      image: container.image,
+    };
+    if (container.cpuPercent !== undefined) ctx.cpu_usage = formatPercent(container.cpuPercent);
+    if (container.memoryUsageBytes !== undefined) ctx.memory_used = formatBytes(container.memoryUsageBytes);
+    if (container.memoryLimitBytes !== undefined) ctx.memory_limit = formatBytes(container.memoryLimitBytes);
+    if (container.memoryPercent !== undefined) ctx.memory_usage = formatPercent(container.memoryPercent);
+    if (container.uptimeSeconds) ctx.uptime = formatUptime(container.uptimeSeconds);
+    if (container.ports?.length) ctx.ports = container.ports.map(p => p.publicPort ? `${p.publicPort}:${p.privatePort}/${p.protocol}` : `${p.privatePort}/${p.protocol}`);
+    if (annotations().length > 0) ctx.user_notes = annotations().join('; ');
+    return ctx;
+  };
+
+  const handleAskAI = () => {
+    aiChatStore.openForTarget('container', resourceId(), {
+      containerName: container.name,
+      ...buildContainerContext(),
+    });
+  };
+
  const writableLayerBytes = createMemo(() => container.writableLayerBytes ?? 0);
  const rootFilesystemBytes = createMemo(() => container.rootFilesystemBytes ?? 0);
  const hasDiskStats = createMemo(() => writableLayerBytes() > 0 || rootFilesystemBytes() > 0);
@@ -1630,6 +1711,72 @@ const DockerContainerRow: Component<{
                </div>
              </div>
                </Show>
+
+              {/* Annotations & Ask AI row */}
+              <Show when={aiEnabled()}>
+                <div class="mt-3 pt-3 border-t border-gray-200 dark:border-gray-700 w-full space-y-2">
+                  <div class="flex items-center gap-1.5">
+                    <span class="text-[10px] font-medium text-gray-500 dark:text-gray-400">AI Context</span>
+                    <Show when={saving()}>
+                      <span class="text-[9px] text-gray-400">saving...</span>
+                    </Show>
+                  </div>
+
+                  {/* Existing annotations */}
+                  <Show when={annotations().length > 0}>
+                    <div class="flex flex-wrap gap-1.5">
+                      <For each={annotations()}>
+                        {(annotation, index) => (
+                          <span class="inline-flex items-center gap-1 px-2 py-1 text-[11px] rounded-md bg-purple-100 text-purple-800 dark:bg-purple-900/40 dark:text-purple-200">
+                            <span class="max-w-[300px] truncate">{annotation}</span>
+                            <button
+                              type="button"
+                              onClick={() => removeAnnotation(index())}
+                              class="ml-0.5 p-0.5 rounded hover:bg-purple-200 dark:hover:bg-purple-800 transition-colors"
+                              title="Remove"
+                            >
+                              <svg class="w-3 h-3" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                                <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M6 18L18 6M6 6l12 12" />
+                              </svg>
+                            </button>
+                          </span>
+                        )}
+                      </For>
+                    </div>
+                  </Show>
+
+                  {/* Add new annotation */}
+                  <div class="flex items-center gap-2">
+                    <input
+                      type="text"
+                      value={newAnnotation()}
+                      onInput={(e) => setNewAnnotation(e.currentTarget.value)}
+                      onKeyDown={handleKeyDown}
+                      placeholder="Add context for AI (press Enter)..."
+                      class="flex-1 px-2 py-1.5 text-[11px] rounded border border-gray-200 dark:border-gray-600 bg-white dark:bg-gray-800 text-gray-700 dark:text-gray-200 placeholder-gray-400 dark:placeholder-gray-500 focus:outline-none focus:ring-1 focus:ring-purple-500 focus:border-purple-500"
+                    />
+                    <button
+                      type="button"
+                      onClick={addAnnotation}
+                      disabled={!newAnnotation().trim()}
+                      class="px-2 py-1.5 text-[11px] rounded border border-purple-300 dark:border-purple-600 text-purple-700 dark:text-purple-300 hover:bg-purple-50 dark:hover:bg-purple-900/30 disabled:opacity-40 disabled:cursor-not-allowed transition-colors"
+                    >
+                      Add
+                    </button>
+                    <button
+                      type="button"
+                      onClick={handleAskAI}
+                      class="flex items-center gap-1.5 px-3 py-1.5 rounded-lg bg-gradient-to-r from-purple-500 to-pink-500 text-white text-[11px] font-medium shadow-sm hover:from-purple-600 hover:to-pink-600 transition-all"
+                      title={`Ask AI about ${container.name}`}
+                    >
+                      <svg class="w-3.5 h-3.5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                        <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9.75 3.104v5.714a2.25 2.25 0 01-.659 1.591L5 14.5M9.75 3.104c-.251.023-.501.05-.75.082m.75-.082a24.301 24.301 0 014.5 0m0 0v5.714c0 .597.237 1.17.659 1.591L19.8 15.3M14.25 3.104c.251.023.501.05.75.082M19.8 15.3l-1.57.393A9.065 9.065 0 0112 15a9.065 9.065 0 00-6.23.693L5 14.5m14.8.8l1.402 1.402c1.232 1.232.65 3.318-1.067 3.611l-2.576.43a18.003 18.003 0 01-5.118 0l-2.576-.43c-1.717-.293-2.299-2.379-1.067-3.611L5 14.5" />
+                      </svg>
+                      Ask AI
+                    </button>
+                  </div>
+                </div>
+              </Show>
              </div>
            </div>
          </td>
--- a/frontend-modern/src/components/Hosts/HostsOverview.tsx
+++ b/frontend-modern/src/components/Hosts/HostsOverview.tsx
@@ -16,6 +16,10 @@ import { StackedMemoryBar } from '@/components/Dashboard/StackedMemoryBar';
 import { EnhancedCPUBar } from '@/components/Dashboard/EnhancedCPUBar';
 import { TemperatureGauge } from '@/components/shared/TemperatureGauge';
 import { useBreakpoint } from '@/hooks/useBreakpoint';
+import { GuestMetadataAPI } from '@/api/guestMetadata';
+import { AIAPI } from '@/api/ai';
+import { aiChatStore } from '@/stores/aiChat';
+import { logger } from '@/utils/logger';

 // Global drawer state to persist across re-renders
 const drawerState = new Map<string, boolean>();
@@ -335,6 +339,76 @@ export const HostsOverview: Component<HostsOverviewProps> = (props) => {
                            // Drawer state
                            const [drawerOpen, setDrawerOpen] = createSignal(drawerState.get(host.id) ?? false);

+                            // AI and annotations state
+                            const [aiEnabled, setAiEnabled] = createSignal(false);
+                            const [annotations, setAnnotations] = createSignal<string[]>([]);
+                            const [newAnnotation, setNewAnnotation] = createSignal('');
+                            const [saving, setSaving] = createSignal(false);
+
+                            // Load AI settings and annotations when drawer opens
+                            createEffect(() => {
+                              if (drawerOpen()) {
+                                AIAPI.getSettings()
+                                  .then((settings) => setAiEnabled(settings.enabled && settings.configured))
+                                  .catch((err) => logger.debug('[HostsOverview] AI settings check failed:', err));
+
+                                GuestMetadataAPI.getMetadata(`host-${host.id}`)
+                                  .then((meta) => {
+                                    if (meta.notes && Array.isArray(meta.notes)) setAnnotations(meta.notes);
+                                  })
+                                  .catch((err) => logger.debug('[HostsOverview] Failed to load annotations:', err));
+                              }
+                            });
+
+                            const saveAnnotations = async (updated: string[]) => {
+                              setSaving(true);
+                              try {
+                                await GuestMetadataAPI.updateMetadata(`host-${host.id}`, { notes: updated });
+                              } catch (err) {
+                                logger.error('[HostsOverview] Failed to save annotations:', err);
+                              } finally {
+                                setSaving(false);
+                              }
+                            };
+
+                            const addAnnotation = () => {
+                              const text = newAnnotation().trim();
+                              if (!text) return;
+                              const updated = [...annotations(), text];
+                              setAnnotations(updated);
+                              setNewAnnotation('');
+                              saveAnnotations(updated);
+                            };
+
+                            const removeAnnotation = (index: number) => {
+                              const updated = annotations().filter((_, i) => i !== index);
+                              setAnnotations(updated);
+                              saveAnnotations(updated);
+                            };
+
+                            const handleKeyDown = (e: KeyboardEvent) => {
+                              if (e.key === 'Enter') {
+                                e.preventDefault();
+                                addAnnotation();
+                              }
+                            };
+
+                            const handleAskAI = () => {
+                              const context: Record<string, unknown> = {
+                                hostName: host.displayName || host.hostname,
+                                hostname: host.hostname,
+                                platform: host.platform,
+                                osName: host.osName,
+                                osVersion: host.osVersion,
+                                cpuUsage: host.cpuUsage ? `${host.cpuUsage.toFixed(1)}%` : undefined,
+                                memoryUsage: host.memory?.usage ? `${host.memory.usage.toFixed(1)}%` : undefined,
+                                uptime: host.uptimeSeconds ? formatUptime(host.uptimeSeconds) : undefined,
+                              };
+                              if (annotations().length > 0) context.user_annotations = annotations();
+
+                              aiChatStore.openForTarget('host', host.id, context);
+                            };
+
                            // Check if we have additional info to show in drawer
                            const hasDrawerContent = createMemo(() => {
                              return (
@@ -687,6 +761,72 @@ export const HostsOverview: Component<HostsOverviewProps> = (props) => {
                                            </For>
                                          </Show>
                                        </div>
+
+                                        {/* AI Context & Ask AI row */}
+                                        <Show when={aiEnabled()}>
+                                          <div class="mt-3 pt-3 border-t border-gray-200 dark:border-gray-700 space-y-2">
+                                            <div class="flex items-center gap-1.5">
+                                              <span class="text-[10px] font-medium text-gray-500 dark:text-gray-400">AI Context</span>
+                                              <Show when={saving()}>
+                                                <span class="text-[9px] text-gray-400">saving...</span>
+                                              </Show>
+                                            </div>
+
+                                            {/* Existing annotations */}
+                                            <Show when={annotations().length > 0}>
+                                              <div class="flex flex-wrap gap-1.5">
+                                                <For each={annotations()}>
+                                                  {(annotation, index) => (
+                                                    <span class="inline-flex items-center gap-1 px-2 py-1 text-[11px] rounded-md bg-purple-100 text-purple-800 dark:bg-purple-900/40 dark:text-purple-200">
+                                                      <span class="max-w-[300px] truncate">{annotation}</span>
+                                                      <button
+                                                        type="button"
+                                                        onClick={() => removeAnnotation(index())}
+                                                        class="ml-0.5 p-0.5 rounded hover:bg-purple-200 dark:hover:bg-purple-800 transition-colors"
+                                                        title="Remove"
+                                                      >
+                                                        <svg class="w-3 h-3" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                                                          <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M6 18L18 6M6 6l12 12" />
+                                                        </svg>
+                                                      </button>
+                                                    </span>
+                                                  )}
+                                                </For>
+                                              </div>
+                                            </Show>
+
+                                            {/* Add new annotation + Ask AI */}
+                                            <div class="flex items-center gap-2">
+                                              <input
+                                                type="text"
+                                                value={newAnnotation()}
+                                                onInput={(e) => setNewAnnotation(e.currentTarget.value)}
+                                                onKeyDown={handleKeyDown}
+                                                placeholder="Add context for AI (press Enter)..."
+                                                class="flex-1 px-2 py-1.5 text-[11px] rounded border border-gray-200 dark:border-gray-600 bg-white dark:bg-gray-800 text-gray-700 dark:text-gray-200 placeholder-gray-400 dark:placeholder-gray-500 focus:outline-none focus:ring-1 focus:ring-purple-500 focus:border-purple-500"
+                                              />
+                                              <button
+                                                type="button"
+                                                onClick={addAnnotation}
+                                                disabled={!newAnnotation().trim()}
+                                                class="px-2 py-1.5 text-[11px] rounded border border-gray-200 dark:border-gray-600 bg-white dark:bg-gray-800 text-gray-700 dark:text-gray-200 hover:bg-gray-50 dark:hover:bg-gray-700 disabled:opacity-50 disabled:cursor-not-allowed transition-colors"
+                                              >
+                                                Add
+                                              </button>
+                                              <button
+                                                type="button"
+                                                onClick={handleAskAI}
+                                                class="flex items-center gap-1.5 px-3 py-1.5 rounded-lg bg-gradient-to-r from-purple-500 to-pink-500 text-white text-[11px] font-medium shadow-sm hover:from-purple-600 hover:to-pink-600 transition-all"
+                                                title={`Ask AI about ${host.displayName || host.hostname}`}
+                                              >
+                                                <svg class="w-3.5 h-3.5" fill="none" stroke="currentColor" viewBox="0 0 24 24">
+                                                  <path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M9.75 3.104v5.714a2.25 2.25 0 01-.659 1.591L5 14.5M9.75 3.104c-.251.023-.501.05-.75.082m.75-.082a24.301 24.301 0 014.5 0m0 0v5.714c0 .597.237 1.17.659 1.591L19.8 15.3M14.25 3.104c.251.023.501.05.75.082M19.8 15.3l-1.57.393A9.065 9.065 0 0112 15a9.065 9.065 0 00-6.23.693L5 14.5m14.8.8l1.402 1.402c1.232 1.232.65 3.318-1.067 3.611l-2.576.43a18.003 18.003 0 01-5.118 0l-2.576-.43c-1.717-.293-2.299-2.379-1.067-3.611L5 14.5" />
+                                                </svg>
+                                                Ask AI
+                                              </button>
+                                            </div>
+                                          </div>
+                                        </Show>
                                      </div>
                                    </td>
                                  </tr>
--- a/frontend-modern/src/components/Settings/AISettings.tsx
+++ b/frontend-modern/src/components/Settings/AISettings.tsx
@@ -0,0 +1,424 @@
+import { Component, Show, createSignal, onMount, For } from 'solid-js';
+import { createStore } from 'solid-js/store';
+import { Card } from '@/components/shared/Card';
+import { SectionHeader } from '@/components/shared/SectionHeader';
+import { Toggle } from '@/components/shared/Toggle';
+import { formField, labelClass, controlClass, formHelpText } from '@/components/shared/Form';
+import { notificationStore } from '@/stores/notifications';
+import { logger } from '@/utils/logger';
+import { AIAPI } from '@/api/ai';
+import type { AISettings as AISettingsType, AIProvider } from '@/types/ai';
+import { PROVIDER_NAMES, PROVIDER_DESCRIPTIONS, DEFAULT_MODELS } from '@/types/ai';
+
+const PROVIDERS: AIProvider[] = ['anthropic', 'openai', 'ollama'];
+
+export const AISettings: Component = () => {
+  const [settings, setSettings] = createSignal<AISettingsType | null>(null);
+  const [loading, setLoading] = createSignal(false);
+  const [saving, setSaving] = createSignal(false);
+  const [testing, setTesting] = createSignal(false);
+
+  const [form, setForm] = createStore({
+    enabled: false,
+    provider: 'anthropic' as AIProvider,
+    apiKey: '',
+    model: '',
+    baseUrl: '',
+    clearApiKey: false,
+    autonomousMode: false,
+  });
+
+  const resetForm = (data: AISettingsType | null) => {
+    if (!data) {
+      setForm({
+        enabled: false,
+        provider: 'anthropic',
+        apiKey: '',
+        model: DEFAULT_MODELS.anthropic,
+        baseUrl: '',
+        clearApiKey: false,
+        autonomousMode: false,
+      });
+      return;
+    }
+
+    setForm({
+      enabled: data.enabled,
+      provider: data.provider,
+      apiKey: '',
+      model: data.model || DEFAULT_MODELS[data.provider],
+      baseUrl: data.base_url || '',
+      clearApiKey: false,
+      autonomousMode: data.autonomous_mode || false,
+    });
+  };
+
+  const loadSettings = async () => {
+    setLoading(true);
+    try {
+      const data = await AIAPI.getSettings();
+      setSettings(data);
+      resetForm(data);
+    } catch (error) {
+      logger.error('[AISettings] Failed to load settings:', error);
+      notificationStore.error('Failed to load AI settings');
+      setSettings(null);
+      resetForm(null);
+    } finally {
+      setLoading(false);
+    }
+  };
+
+  onMount(() => {
+    loadSettings();
+  });
+
+  const handleProviderChange = (provider: AIProvider) => {
+    setForm('provider', provider);
+    // Update model to default for new provider if current model doesn't look like it belongs
+    const currentModel = form.model;
+    if (!currentModel || currentModel === DEFAULT_MODELS[settings()?.provider || 'anthropic']) {
+      setForm('model', DEFAULT_MODELS[provider]);
+    }
+  };
+
+  const handleSave = async (event?: Event) => {
+    event?.preventDefault();
+
+    setSaving(true);
+    try {
+      const payload: Record<string, unknown> = {
+        provider: form.provider,
+        model: form.model.trim(),
+      };
+
+      // Only include base_url if it's set or if provider is ollama
+      if (form.baseUrl.trim() || form.provider === 'ollama') {
+        payload.base_url = form.baseUrl.trim();
+      }
+
+      // Handle API key
+      if (form.apiKey.trim() !== '') {
+        payload.api_key = form.apiKey.trim();
+      } else if (form.clearApiKey) {
+        payload.api_key = '';
+      }
+
+      // Only include enabled if we're toggling it
+      if (form.enabled !== settings()?.enabled) {
+        payload.enabled = form.enabled;
+      }
+
+      // Include autonomous mode if changed
+      if (form.autonomousMode !== settings()?.autonomous_mode) {
+        payload.autonomous_mode = form.autonomousMode;
+      }
+
+      const updated = await AIAPI.updateSettings(payload);
+      setSettings(updated);
+      resetForm(updated);
+      notificationStore.success('AI settings saved');
+    } catch (error) {
+      logger.error('[AISettings] Failed to save settings:', error);
+      const message = error instanceof Error ? error.message : 'Failed to save AI settings';
+      notificationStore.error(message);
+    } finally {
+      setSaving(false);
+    }
+  };
+
+  const handleTest = async () => {
+    setTesting(true);
+    try {
+      const result = await AIAPI.testConnection();
+      if (result.success) {
+        notificationStore.success(result.message);
+      } else {
+        notificationStore.error(result.message);
+      }
+    } catch (error) {
+      logger.error('[AISettings] Test failed:', error);
+      const message = error instanceof Error ? error.message : 'Connection test failed';
+      notificationStore.error(message);
+    } finally {
+      setTesting(false);
+    }
+  };
+
+  const needsApiKey = () => form.provider !== 'ollama';
+  const showBaseUrl = () => form.provider === 'ollama' || form.provider === 'openai';
+
+  return (
+    <Card
+      padding="none"
+      class="overflow-hidden border border-gray-200 dark:border-gray-700"
+      border={false}
+    >
+      <div class="bg-gradient-to-r from-purple-50 to-pink-50 dark:from-purple-900/20 dark:to-pink-900/20 px-6 py-4 border-b border-gray-200 dark:border-gray-700">
+        <div class="flex items-center gap-3">
+          <div class="p-2 bg-purple-100 dark:bg-purple-900/40 rounded-lg">
+            <svg
+              class="w-5 h-5 text-purple-600 dark:text-purple-300"
+              fill="none"
+              stroke="currentColor"
+              viewBox="0 0 24 24"
+            >
+              <path
+                stroke-linecap="round"
+                stroke-linejoin="round"
+                stroke-width="1.8"
+                d="M9.75 3.104v5.714a2.25 2.25 0 01-.659 1.591L5 14.5M9.75 3.104c-.251.023-.501.05-.75.082m.75-.082a24.301 24.301 0 014.5 0m0 0v5.714c0 .597.237 1.17.659 1.591L19.8 15.3M14.25 3.104c.251.023.501.05.75.082M19.8 15.3l-1.57.393A9.065 9.065 0 0112 15a9.065 9.065 0 00-6.23.693L5 14.5m14.8.8l1.402 1.402c1.232 1.232.65 3.318-1.067 3.611l-2.576.43a18.003 18.003 0 01-5.118 0l-2.576-.43c-1.717-.293-2.299-2.379-1.067-3.611L5 14.5"
+              />
+            </svg>
+          </div>
+          <SectionHeader
+            title="AI Assistant"
+            description="Configure AI-powered infrastructure analysis"
+            size="sm"
+            class="flex-1"
+          />
+          <Toggle
+            checked={form.enabled}
+            onChange={(event) => {
+              setForm('enabled', event.currentTarget.checked);
+            }}
+            disabled={loading() || saving()}
+            containerClass="items-center gap-2"
+            label={
+              <span class="text-xs font-medium text-gray-600 dark:text-gray-300">
+                {form.enabled ? 'Enabled' : 'Disabled'}
+              </span>
+            }
+          />
+        </div>
+      </div>
+
+      <form class="p-6 space-y-5" onSubmit={handleSave}>
+        <div class="bg-purple-50 dark:bg-purple-900/30 border border-purple-200 dark:border-purple-800 rounded-lg p-3 text-xs text-purple-800 dark:text-purple-200">
+          <p class="font-medium mb-1">AI Assistant helps you:</p>
+          <ul class="space-y-0.5 list-disc pl-4">
+            <li>Diagnose infrastructure issues with context-aware analysis</li>
+            <li>Get remediation suggestions based on your specific environment</li>
+            <li>Understand alerts and metrics with plain-language explanations</li>
+          </ul>
+        </div>
+
+        <Show when={loading()}>
+          <div class="flex items-center gap-3 text-sm text-gray-600 dark:text-gray-300">
+            <span class="h-4 w-4 border-2 border-current border-t-transparent rounded-full animate-spin" />
+            Loading AI settings...
+          </div>
+        </Show>
+
+        <Show when={!loading()}>
+          <div class="space-y-4">
+            {/* Provider Selection */}
+            <div class={formField}>
+              <label class={labelClass()}>AI Provider</label>
+              <div class="grid grid-cols-3 gap-3">
+                <For each={PROVIDERS}>
+                  {(provider) => (
+                    <button
+                      type="button"
+                      class={`p-3 rounded-lg border-2 text-left transition-all ${
+                        form.provider === provider
+                          ? 'border-purple-500 bg-purple-50 dark:bg-purple-900/30'
+                          : 'border-gray-200 dark:border-gray-700 hover:border-gray-300 dark:hover:border-gray-600'
+                      }`}
+                      onClick={() => handleProviderChange(provider)}
+                      disabled={saving()}
+                    >
+                      <div class="font-medium text-sm text-gray-900 dark:text-gray-100">
+                        {PROVIDER_NAMES[provider]}
+                      </div>
+                      <div class="text-xs text-gray-500 dark:text-gray-400 mt-0.5">
+                        {PROVIDER_DESCRIPTIONS[provider]}
+                      </div>
+                    </button>
+                  )}
+                </For>
+              </div>
+            </div>
+
+            {/* API Key - not shown for Ollama */}
+            <Show when={needsApiKey()}>
+              <div class={formField}>
+                <div class="flex items-center justify-between">
+                  <label class={labelClass('mb-0')}>API Key</label>
+                  <Show when={settings()?.api_key_set}>
+                    <button
+                      type="button"
+                      class="text-xs text-purple-600 hover:underline dark:text-purple-300"
+                      onClick={() => {
+                        if (!saving()) {
+                          setForm('apiKey', '');
+                          setForm('clearApiKey', true);
+                          notificationStore.info('API key will be cleared on save', 2500);
+                        }
+                      }}
+                      disabled={saving()}
+                    >
+                      Clear stored key
+                    </button>
+                  </Show>
+                </div>
+                <input
+                  type="password"
+                  value={form.apiKey}
+                  onInput={(event) => {
+                    setForm('apiKey', event.currentTarget.value);
+                    if (event.currentTarget.value.trim() !== '') {
+                      setForm('clearApiKey', false);
+                    }
+                  }}
+                  placeholder={
+                    settings()?.api_key_set
+                      ? '•••••••• (leave blank to keep existing)'
+                      : `Enter ${PROVIDER_NAMES[form.provider]} API key`
+                  }
+                  class={controlClass()}
+                  disabled={saving()}
+                />
+                <p class={formHelpText}>
+                  {form.provider === 'anthropic'
+                    ? 'Get your API key from console.anthropic.com'
+                    : 'Get your API key from platform.openai.com'}
+                </p>
+              </div>
+            </Show>
+
+            {/* Model */}
+            <div class={formField}>
+              <label class={labelClass()}>Model</label>
+              <input
+                type="text"
+                value={form.model}
+                onInput={(event) => setForm('model', event.currentTarget.value)}
+                placeholder={DEFAULT_MODELS[form.provider]}
+                class={controlClass()}
+                disabled={saving()}
+              />
+              <p class={formHelpText}>
+                {form.provider === 'anthropic'
+                  ? 'e.g., claude-opus-4-5-20251101, claude-sonnet-4-20250514'
+                  : form.provider === 'openai'
+                    ? 'e.g., gpt-4o, gpt-4-turbo'
+                    : 'e.g., llama3, mixtral, codellama'}
+              </p>
+            </div>
+
+            {/* Base URL - shown for Ollama (required) and OpenAI (optional) */}
+            <Show when={showBaseUrl()}>
+              <div class={formField}>
+                <label class={labelClass()}>
+                  {form.provider === 'ollama' ? 'Ollama Server URL' : 'API Base URL (optional)'}
+                </label>
+                <input
+                  type="url"
+                  value={form.baseUrl}
+                  onInput={(event) => setForm('baseUrl', event.currentTarget.value)}
+                  placeholder={
+                    form.provider === 'ollama'
+                      ? 'http://localhost:11434'
+                      : 'https://api.openai.com/v1'
+                  }
+                  class={controlClass()}
+                  disabled={saving()}
+                />
+                <p class={formHelpText}>
+                  {form.provider === 'ollama'
+                    ? 'URL where your Ollama server is running'
+                    : 'Custom endpoint for Azure OpenAI or compatible APIs'}
+                </p>
+              </div>
+            </Show>
+
+            {/* Autonomous Mode */}
+            <div class={`${formField} p-4 rounded-lg border ${form.autonomousMode ? 'bg-amber-50 dark:bg-amber-900/20 border-amber-200 dark:border-amber-800' : 'bg-gray-50 dark:bg-gray-800/50 border-gray-200 dark:border-gray-700'}`}>
+              <div class="flex items-start justify-between gap-4">
+                <div class="flex-1">
+                  <label class={`${labelClass()} flex items-center gap-2`}>
+                    Autonomous Mode
+                    <Show when={form.autonomousMode}>
+                      <span class="px-1.5 py-0.5 text-[10px] font-semibold bg-amber-200 dark:bg-amber-800 text-amber-800 dark:text-amber-200 rounded">
+                        ENABLED
+                      </span>
+                    </Show>
+                  </label>
+                  <p class="text-xs text-gray-600 dark:text-gray-400 mt-1">
+                    {form.autonomousMode
+                      ? 'AI will execute all commands without asking for approval. Only enable if you trust your configured model.'
+                      : 'AI will ask for approval before running commands that modify your system. Read-only commands (like df, ps, docker stats) run automatically.'}
+                  </p>
+                </div>
+                <Toggle
+                  checked={form.autonomousMode}
+                  onChange={(event) => setForm('autonomousMode', event.currentTarget.checked)}
+                  disabled={saving()}
+                />
+              </div>
+            </div>
+
+          </div>
+
+          {/* Status indicator */}
+          <Show when={settings()}>
+            <div
+              class={`flex items-center gap-2 p-3 rounded-lg ${
+                settings()?.configured
+                  ? 'bg-green-50 dark:bg-green-900/30 text-green-800 dark:text-green-200'
+                  : 'bg-amber-50 dark:bg-amber-900/30 text-amber-800 dark:text-amber-200'
+              }`}
+            >
+              <div
+                class={`w-2 h-2 rounded-full ${
+                  settings()?.configured ? 'bg-green-500' : 'bg-amber-500'
+                }`}
+              />
+              <span class="text-xs font-medium">
+                {settings()?.configured
+                  ? `Ready to use with ${settings()?.model}`
+                  : needsApiKey()
+                    ? 'API key required to enable AI features'
+                    : 'Configure Ollama server URL to enable AI features'}
+              </span>
+            </div>
+          </Show>
+
+          {/* Actions */}
+          <div class="flex flex-wrap items-center justify-between gap-3 pt-4">
+            <Show when={settings()?.api_key_set}>
+              <button
+                type="button"
+                class="px-4 py-2 text-sm border border-purple-300 dark:border-purple-600 text-purple-700 dark:text-purple-300 rounded-md hover:bg-purple-50 dark:hover:bg-purple-900/30 disabled:opacity-50 disabled:cursor-not-allowed"
+                onClick={handleTest}
+                disabled={testing() || saving() || loading()}
+              >
+                {testing() ? 'Testing...' : 'Test Connection'}
+              </button>
+            </Show>
+            <div class="flex gap-3 ml-auto">
+              <button
+                type="button"
+                class="px-4 py-2 border border-gray-300 dark:border-gray-600 text-gray-700 dark:text-gray-300 rounded-md hover:bg-gray-50 dark:hover:bg-gray-700"
+                onClick={() => resetForm(settings())}
+                disabled={saving() || loading()}
+              >
+                Reset
+              </button>
+              <button
+                type="submit"
+                class="px-4 py-2 bg-purple-600 text-white rounded-md hover:bg-purple-700 disabled:opacity-50 disabled:cursor-not-allowed"
+                disabled={saving() || loading()}
+              >
+                {saving() ? 'Saving...' : 'Save changes'}
+              </button>
+            </div>
+          </div>
+        </Show>
+      </form>
+    </Card>
+  );
+};
+
+export default AISettings;
--- a/frontend-modern/src/components/Settings/Settings.tsx
+++ b/frontend-modern/src/components/Settings/Settings.tsx
@@ -27,6 +27,7 @@ import { ChangePasswordModal } from './ChangePasswordModal';
 import { UnifiedAgents } from './UnifiedAgents';
 import APITokenManager from './APITokenManager';
 import { OIDCPanel } from './OIDCPanel';
+import { AISettings } from './AISettings';
 import { QuickSecuritySetup } from './QuickSecuritySetup';
 import { SecurityPostureSummary } from './SecurityPostureSummary';
 import {
@@ -57,6 +58,7 @@ import Monitor from 'lucide-solid/icons/monitor';
 import Sliders from 'lucide-solid/icons/sliders-horizontal';
 import RefreshCw from 'lucide-solid/icons/refresh-cw';
 import Clock from 'lucide-solid/icons/clock';
+import Sparkles from 'lucide-solid/icons/sparkles';
 import { ProxmoxIcon } from '@/components/icons/ProxmoxIcon';
 import BadgeCheck from 'lucide-solid/icons/badge-check';
 import type { NodeConfig } from '@/types/nodes';
@@ -324,6 +326,7 @@ type SettingsTab =
  | 'system-network'
  | 'system-updates'
  | 'system-backups'
+  | 'system-ai'
  | 'api'
  | 'security-overview'
  | 'security-auth'
@@ -369,6 +372,10 @@ const SETTINGS_HEADER_META: Record<SettingsTab, { title: string; description: st
    title: 'Backup Polling',
    description: 'Control how often Pulse queries Proxmox for backup tasks and snapshots.',
  },
+  'system-ai': {
+    title: 'AI Assistant',
+    description: 'Configure AI-powered infrastructure analysis and remediation suggestions.',
+  },
  api: {
    title: 'API access',
    description:
@@ -1281,6 +1288,12 @@ const Settings: Component<SettingsProps> = (props) => {
            icon: Clock,
            iconProps: { strokeWidth: 2 },
          },
+          {
+            id: 'system-ai',
+            label: 'AI Assistant',
+            icon: Sparkles,
+            iconProps: { strokeWidth: 2 },
+          },
        ],
      },
      {
@@ -4915,6 +4928,11 @@ const Settings: Component<SettingsProps> = (props) => {
                </div>
              </Show>

+              {/* AI Assistant Tab */}
+              <Show when={activeTab() === 'system-ai'}>
+                <AISettings />
+              </Show>
+
              {/* API Access */}
              <Show when={activeTab() === 'api'}>
                <div class="space-y-6">
--- a/frontend-modern/src/stores/aiChat.ts
+++ b/frontend-modern/src/stores/aiChat.ts
@@ -0,0 +1,192 @@
+import { createSignal } from 'solid-js';
+
+interface AIChatContext {
+  targetType?: string;
+  targetId?: string;
+  context?: Record<string, unknown>;
+  initialPrompt?: string;
+}
+
+// A single context item that can be accumulated
+interface ContextItem {
+  id: string; // unique identifier (e.g., "vm-delly-101")
+  type: string; // "vm", "container", "storage", "node", etc.
+  name: string; // display name
+  data: Record<string, unknown>; // the actual context data
+  addedAt: Date;
+}
+
+// Message type for persisted conversation
+interface Message {
+  id: string;
+  role: 'user' | 'assistant';
+  content: string;
+  timestamp: Date;
+  model?: string;
+  tokens?: { input: number; output: number };
+  toolCalls?: Array<{
+    name: string;
+    input: string;
+    output: string;
+    success: boolean;
+  }>;
+}
+
+// Global state for the AI chat drawer
+const [isAIChatOpen, setIsAIChatOpen] = createSignal(false);
+const [aiChatContext, setAIChatContext] = createSignal<AIChatContext>({});
+const [contextItems, setContextItems] = createSignal<ContextItem[]>([]);
+const [messages, setMessages] = createSignal<Message[]>([]);
+const [aiEnabled, setAiEnabled] = createSignal<boolean | null>(null); // null = not checked yet
+
+export const aiChatStore = {
+  // Check if chat is open
+  get isOpen() {
+    return isAIChatOpen();
+  },
+
+  // Get current context (legacy single-item)
+  get context() {
+    return aiChatContext();
+  },
+
+  // Get all accumulated context items
+  get contextItems() {
+    return contextItems();
+  },
+
+  // Get messages (for persistence)
+  get messages() {
+    return messages();
+  },
+
+  // Get AI enabled state
+  get enabled() {
+    return aiEnabled();
+  },
+
+  // Check if a specific item is in context
+  hasContextItem(id: string) {
+    return contextItems().some(item => item.id === id);
+  },
+
+  // Set AI enabled state (called from settings check)
+  setEnabled(enabled: boolean) {
+    setAiEnabled(enabled);
+  },
+
+  // Set messages (for persistence from AIChat component)
+  setMessages(msgs: Message[]) {
+    setMessages(msgs);
+  },
+
+  // Toggle the AI chat panel
+  toggle() {
+    setIsAIChatOpen(!isAIChatOpen());
+  },
+
+  // Open the AI chat with optional context
+  open(context?: AIChatContext) {
+    if (context) {
+      setAIChatContext(context);
+    }
+    setIsAIChatOpen(true);
+  },
+
+  // Close the AI chat
+  close() {
+    setIsAIChatOpen(false);
+    // Keep context and messages for when user reopens
+  },
+
+  // Update context without opening (for navigation-based context changes)
+  setContext(context: AIChatContext) {
+    setAIChatContext(context);
+  },
+
+  // Clear single-item context (legacy)
+  clearContext() {
+    setAIChatContext({});
+  },
+
+  // Add an item to the context (accumulative)
+  addContextItem(type: string, id: string, name: string, data: Record<string, unknown>) {
+    setContextItems(prev => {
+      // Don't add duplicates
+      if (prev.some(item => item.id === id)) {
+        // Update existing item with new data
+        return prev.map(item =>
+          item.id === id
+            ? { ...item, data, addedAt: new Date() }
+            : item
+        );
+      }
+      return [...prev, { id, type, name, data, addedAt: new Date() }];
+    });
+    // Also update legacy context to point to most recently added
+    setAIChatContext({
+      targetType: type,
+      targetId: id,
+      context: data,
+    });
+  },
+
+  // Remove an item from context
+  removeContextItem(id: string) {
+    setContextItems(prev => prev.filter(item => item.id !== id));
+    // Update legacy context if we removed the current one
+    const current = aiChatContext();
+    if (current.targetId === id) {
+      const remaining = contextItems().filter(item => item.id !== id);
+      if (remaining.length > 0) {
+        const last = remaining[remaining.length - 1];
+        setAIChatContext({
+          targetType: last.type,
+          targetId: last.id,
+          context: last.data,
+        });
+      } else {
+        setAIChatContext({});
+      }
+    }
+  },
+
+  // Clear all context items
+  clearAllContext() {
+    setContextItems([]);
+    setAIChatContext({});
+  },
+
+  // Clear conversation (start fresh)
+  clearConversation() {
+    setMessages([]);
+  },
+
+  // Convenience method to update context for a specific target (host, VM, container, etc.)
+  // This is called when user selects/views a specific resource
+  setTargetContext(targetType: string, targetId: string, additionalContext?: Record<string, unknown>) {
+    // Use addContextItem instead of replacing
+    const name = (additionalContext?.guestName as string) ||
+                 (additionalContext?.name as string) ||
+                 targetId;
+    this.addContextItem(targetType, targetId, name, additionalContext || {});
+  },
+
+  // Open for a specific target - opens the panel and adds to context
+  openForTarget(targetType: string, targetId: string, additionalContext?: Record<string, unknown>) {
+    const name = (additionalContext?.guestName as string) ||
+                 (additionalContext?.name as string) ||
+                 targetId;
+    this.addContextItem(targetType, targetId, name, additionalContext || {});
+    setIsAIChatOpen(true);
+  },
+
+  // Open with a pre-filled prompt
+  openWithPrompt(prompt: string, context?: Omit<AIChatContext, 'initialPrompt'>) {
+    setAIChatContext({
+      ...context,
+      initialPrompt: prompt,
+    });
+    setIsAIChatOpen(true);
+  },
+};
--- a/frontend-modern/src/types/ai.ts
+++ b/frontend-modern/src/types/ai.ts
@@ -0,0 +1,115 @@
+// AI feature types
+
+export type AIProvider = 'anthropic' | 'openai' | 'ollama';
+
+export interface AISettings {
+  enabled: boolean;
+  provider: AIProvider;
+  api_key_set: boolean; // API key is never exposed, just whether it's set
+  model: string;
+  base_url?: string;
+  configured: boolean; // true if AI is ready to use
+  autonomous_mode: boolean; // true if AI can execute commands without approval
+  custom_context: string; // user-provided infrastructure context
+}
+
+export interface AISettingsUpdateRequest {
+  enabled?: boolean;
+  provider?: AIProvider;
+  api_key?: string; // empty string clears, undefined preserves
+  model?: string;
+  base_url?: string;
+  autonomous_mode?: boolean;
+  custom_context?: string; // user-provided infrastructure context
+}
+
+export interface AITestResult {
+  success: boolean;
+  message: string;
+  model?: string;
+}
+
+// Default models for each provider
+export const DEFAULT_MODELS: Record<AIProvider, string> = {
+  anthropic: 'claude-opus-4-5-20251101',
+  openai: 'gpt-4o',
+  ollama: 'llama3',
+};
+
+// Provider display names
+export const PROVIDER_NAMES: Record<AIProvider, string> = {
+  anthropic: 'Anthropic',
+  openai: 'OpenAI',
+  ollama: 'Ollama',
+};
+
+// Provider descriptions
+export const PROVIDER_DESCRIPTIONS: Record<AIProvider, string> = {
+  anthropic: 'Claude models from Anthropic',
+  openai: 'GPT models from OpenAI',
+  ollama: 'Local models via Ollama',
+};
+
+// Conversation history for multi-turn chats
+export interface AIConversationMessage {
+  role: 'user' | 'assistant';
+  content: string;
+}
+
+// AI Execute request/response types
+export interface AIExecuteRequest {
+  prompt: string;
+  target_type?: string; // "host", "container", "vm", "node"
+  target_id?: string;
+  context?: Record<string, unknown>;
+  history?: AIConversationMessage[]; // Previous conversation messages
+}
+
+// Tool execution info
+export interface AIToolExecution {
+  name: string;      // "run_command", "read_file"
+  input: string;     // The command or file path
+  output: string;    // Result of execution
+  success: boolean;
+}
+
+export interface AIExecuteResponse {
+  content: string;
+  model: string;
+  input_tokens: number;
+  output_tokens: number;
+  tool_calls?: AIToolExecution[]; // Commands that were executed
+}
+
+// Streaming event types
+export type AIStreamEventType = 'tool_start' | 'tool_end' | 'content' | 'done' | 'error' | 'complete' | 'approval_needed';
+
+export interface AIStreamToolStartData {
+  name: string;
+  input: string;
+}
+
+export interface AIStreamToolEndData {
+  name: string;
+  output: string;
+  success: boolean;
+}
+
+export interface AIStreamApprovalNeededData {
+  command: string;
+  tool_id: string;
+  tool_name: string;
+  run_on_host: boolean;
+}
+
+export interface AIStreamEvent {
+  type: AIStreamEventType;
+  data?: string | AIStreamToolStartData | AIStreamToolEndData | AIStreamCompleteData | AIStreamApprovalNeededData;
+}
+
+export interface AIStreamCompleteData {
+  model: string;
+  input_tokens: number;
+  output_tokens: number;
+  tool_calls?: AIToolExecution[];
+}
--- a/frontend-modern/tailwind.config.js
+++ b/frontend-modern/tailwind.config.js
@@ -1,3 +1,5 @@
+import typography from '@tailwindcss/typography';
+
 export default {
  content: [
    "./index.html",
@@ -23,5 +25,5 @@ export default {
      }
    },
  },
-  plugins: [],
+  plugins: [typography],
 }
--- a/frontend-modern/vite.config.ts
+++ b/frontend-modern/vite.config.ts
@@ -52,6 +52,41 @@ export default defineConfig({
        ws: true,
        changeOrigin: true,
      },
+      '/api/ai/execute/stream': {
+        target: backendUrl,
+        changeOrigin: true,
+        // SSE requires special handling to prevent proxy timeouts
+        // Set timeout to 10 minutes (600000ms) for long-running AI requests
+        timeout: 600000,
+        proxyTimeout: 600000,
+        configure: (proxy, _options) => {
+          // Set proxy-level timeouts
+          proxy.on('proxyReq', (proxyReq, req, res) => {
+            // Disable socket timeouts for SSE
+            req.socket.setTimeout(0);
+            req.socket.setNoDelay(true);
+            req.socket.setKeepAlive(true);
+            // Also set on the proxy request
+            proxyReq.socket?.setTimeout(0);
+          });
+          proxy.on('proxyRes', (proxyRes, req, res) => {
+            // Disable response socket timeout
+            res.socket?.setTimeout(0);
+            res.socket?.setNoDelay(true);
+            res.socket?.setKeepAlive(true);
+            // Also disable on proxy response socket
+            proxyRes.socket?.setTimeout(0);
+          });
+          proxy.on('error', (err, req, res) => {
+            console.error('[SSE Proxy Error]', err.message);
+          });
+        },
+      },
+      '/api/agent/ws': {
+        target: backendWsUrl,
+        ws: true,
+        changeOrigin: true,
+      },
      '/api': {
        target: backendUrl,
        changeOrigin: true,
--- a/internal/agentexec/policy.go
+++ b/internal/agentexec/policy.go
@@ -0,0 +1,249 @@
+package agentexec
+
+import (
+	"regexp"
+	"strings"
+)
+
+// CommandPolicy defines what commands are allowed, blocked, or require approval
+type CommandPolicy struct {
+	// AutoApprove patterns - commands matching these are automatically allowed
+	AutoApprove []string
+
+	// RequireApproval patterns - commands matching these need user approval
+	RequireApproval []string
+
+	// Blocked patterns - commands matching these are never allowed
+	Blocked []string
+
+	// compiled regex patterns
+	autoApproveRe     []*regexp.Regexp
+	requireApprovalRe []*regexp.Regexp
+	blockedRe         []*regexp.Regexp
+}
+
+// DefaultPolicy returns a sensible default command policy
+func DefaultPolicy() *CommandPolicy {
+	p := &CommandPolicy{
+		AutoApprove: []string{
+			// System inspection
+			`^ps\s`,
+			`^top\s+-bn`,
+			`^df\s`,
+			`^free\s`,
+			`^uptime$`,
+			`^hostname$`,
+			`^uname\s`,
+			`^cat\s+/proc/`,
+			`^cat\s+/etc/os-release`,
+			`^lsof\s`,
+			`^netstat\s`,
+			`^ss\s`,
+			`^ip\s+(addr|route|link)`,
+			`^ifconfig`,
+			`^w$`,
+			`^who$`,
+			`^last\s`,
+
+			// Log reading (read-only)
+			`^cat\s+/var/log/`,
+			`^tail\s+.*(/var/log/|/log/)`,
+			`^head\s+.*(/var/log/|/log/)`,
+			`^grep\s+.*(/var/log/|/log/)`,
+			`^journalctl\s`,
+
+			// Service status (read-only)
+			`^systemctl\s+status\s`,
+			`^systemctl\s+is-active\s`,
+			`^systemctl\s+is-enabled\s`,
+			`^systemctl\s+list-units`,
+			`^service\s+\S+\s+status`,
+
+			// Docker inspection (read-only)
+			`^docker\s+ps`,
+			`^docker\s+logs\s`,
+			`^docker\s+inspect\s`,
+			`^docker\s+stats\s+--no-stream`,
+			`^docker\s+top\s`,
+			`^docker\s+images`,
+			`^docker\s+network\s+ls`,
+			`^docker\s+volume\s+ls`,
+
+			// Proxmox inspection (read-only)
+			`^pct\s+list`,
+			`^pct\s+config\s`,
+			`^pct\s+status\s`,
+			`^qm\s+list`,
+			`^qm\s+config\s`,
+			`^qm\s+status\s`,
+			`^pvesh\s+get\s`,
+
+			// Disk/storage info
+			`^lsblk`,
+			`^blkid`,
+			`^fdisk\s+-l`,
+			`^du\s`,
+			`^ls\s`,
+			`^stat\s`,
+			`^file\s`,
+			`^find\s+/.*-size`,  // Find large files
+			`^find\s+/.*-mtime`, // Find by modification time
+			`^find\s+/.*-type`,  // Find by type
+
+			// Memory inspection
+			`^vmstat`,
+			`^sar\s`,
+			`^iostat`,
+			`^mpstat`,
+
+			// Docker inspection (read-only)
+			`^docker\s+system\s+df`,
+
+			// APT inspection (read-only)
+			`^apt\s+list`,
+			`^apt-cache\s`,
+			`^dpkg\s+-l`,
+			`^dpkg\s+--list`,
+		},
+
+		RequireApproval: []string{
+			// Service control
+			`^systemctl\s+(restart|stop|start|reload)\s`,
+			`^service\s+\S+\s+(restart|stop|start|reload)`,
+
+			// Docker control
+			`^docker\s+(restart|stop|start|kill)\s`,
+			`^docker\s+exec\s`,
+			`^docker\s+rm\s`,
+
+			// Process control
+			`^kill\s`,
+			`^pkill\s`,
+			`^killall\s`,
+
+			// Package management
+			`^apt\s`,
+			`^apt-get\s`,
+			`^yum\s`,
+			`^dnf\s`,
+			`^pacman\s`,
+
+			// Proxmox control
+			`^pct\s+(start|stop|shutdown|reboot|resize|set)\s`,
+			`^qm\s+(start|stop|shutdown|reboot|reset|resize|set)\s`,
+		},
+
+		Blocked: []string{
+			// Destructive filesystem operations
+			`rm\s+-rf\s+/`,
+			`rm\s+--no-preserve-root`,
+			`mkfs`,
+			`dd\s+.*of=/dev/`,
+			`>\s*/dev/sd`,
+			`>\s*/dev/nvme`,
+
+			// System destruction
+			`shutdown`,
+			`reboot`,
+			`init\s+0`,
+			`poweroff`,
+			`halt`,
+
+			// Dangerous permissions
+			`chmod\s+777`,
+			`chmod\s+-R\s+777`,
+			`chown\s+-R\s+.*:.*\s+/`,
+
+			// Remote code execution
+			`curl.*\|\s*(ba)?sh`,
+			`wget.*\|\s*(ba)?sh`,
+			`bash\s+-c\s+.*curl`,
+			`bash\s+-c\s+.*wget`,
+
+			// Crypto mining indicators
+			`xmrig`,
+			`minerd`,
+			`cpuminer`,
+
+			// Fork bomb patterns
+			`:\(\)\s*{\s*:\s*\|\s*:`,
+
+			// Clear system logs
+			`>\s*/var/log/`,
+			`truncate.*--size.*0.*/var/log/`,
+		},
+	}
+
+	p.compile()
+	return p
+}
+
+func (p *CommandPolicy) compile() {
+	p.autoApproveRe = compilePatterns(p.AutoApprove)
+	p.requireApprovalRe = compilePatterns(p.RequireApproval)
+	p.blockedRe = compilePatterns(p.Blocked)
+}
+
+func compilePatterns(patterns []string) []*regexp.Regexp {
+	result := make([]*regexp.Regexp, 0, len(patterns))
+	for _, pattern := range patterns {
+		re, err := regexp.Compile(pattern)
+		if err == nil {
+			result = append(result, re)
+		}
+	}
+	return result
+}
+
+// PolicyDecision represents the policy decision for a command
+type PolicyDecision string
+
+const (
+	PolicyAllow           PolicyDecision = "allow"
+	PolicyRequireApproval PolicyDecision = "require_approval"
+	PolicyBlock           PolicyDecision = "block"
+)
+
+// Evaluate checks a command against the policy
+func (p *CommandPolicy) Evaluate(command string) PolicyDecision {
+	command = strings.TrimSpace(command)
+
+	// Check blocked first (highest priority)
+	for _, re := range p.blockedRe {
+		if re.MatchString(command) {
+			return PolicyBlock
+		}
+	}
+
+	// Check require approval
+	for _, re := range p.requireApprovalRe {
+		if re.MatchString(command) {
+			return PolicyRequireApproval
+		}
+	}
+
+	// Check auto-approve
+	for _, re := range p.autoApproveRe {
+		if re.MatchString(command) {
+			return PolicyAllow
+		}
+	}
+
+	// Default: require approval for unknown commands
+	return PolicyRequireApproval
+}
+
+// IsBlocked returns true if a command is blocked
+func (p *CommandPolicy) IsBlocked(command string) bool {
+	return p.Evaluate(command) == PolicyBlock
+}
+
+// NeedsApproval returns true if a command needs user approval
+func (p *CommandPolicy) NeedsApproval(command string) bool {
+	return p.Evaluate(command) == PolicyRequireApproval
+}
+
+// IsAutoApproved returns true if a command can run without approval
+func (p *CommandPolicy) IsAutoApproved(command string) bool {
+	return p.Evaluate(command) == PolicyAllow
+}
--- a/internal/agentexec/server.go
+++ b/internal/agentexec/server.go
@@ -0,0 +1,360 @@
+package agentexec
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"sync"
+	"time"
+
+	"github.com/gorilla/websocket"
+	"github.com/rs/zerolog/log"
+)
+
+var upgrader = websocket.Upgrader{
+	ReadBufferSize:  1024,
+	WriteBufferSize: 1024,
+	CheckOrigin: func(r *http.Request) bool {
+		return true // Agents can connect from anywhere
+	},
+}
+
+// Server manages WebSocket connections from agents
+type Server struct {
+	mu            sync.RWMutex
+	agents        map[string]*agentConn // agentID -> connection
+	pendingReqs   map[string]chan CommandResultPayload // requestID -> response channel
+	validateToken func(token string) bool
+}
+
+type agentConn struct {
+	conn      *websocket.Conn
+	agent     ConnectedAgent
+	writeMu   sync.Mutex
+	done      chan struct{}
+}
+
+// NewServer creates a new agent execution server
+func NewServer(validateToken func(token string) bool) *Server {
+	return &Server{
+		agents:        make(map[string]*agentConn),
+		pendingReqs:   make(map[string]chan CommandResultPayload),
+		validateToken: validateToken,
+	}
+}
+
+// HandleWebSocket handles incoming WebSocket connections from agents
+func (s *Server) HandleWebSocket(w http.ResponseWriter, r *http.Request) {
+	conn, err := upgrader.Upgrade(w, r, nil)
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to upgrade WebSocket connection")
+		return
+	}
+
+	// Read first message (must be agent_register)
+	conn.SetReadDeadline(time.Now().Add(30 * time.Second))
+	_, msgBytes, err := conn.ReadMessage()
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to read registration message")
+		conn.Close()
+		return
+	}
+
+	var msg Message
+	if err := json.Unmarshal(msgBytes, &msg); err != nil {
+		log.Error().Err(err).Msg("Failed to parse registration message")
+		conn.Close()
+		return
+	}
+
+	if msg.Type != MsgTypeAgentRegister {
+		log.Error().Str("type", string(msg.Type)).Msg("First message must be agent_register")
+		conn.Close()
+		return
+	}
+
+	// Parse registration payload
+	payloadBytes, err := json.Marshal(msg.Payload)
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to marshal registration payload")
+		conn.Close()
+		return
+	}
+
+	var reg AgentRegisterPayload
+	if err := json.Unmarshal(payloadBytes, &reg); err != nil {
+		log.Error().Err(err).Msg("Failed to parse registration payload")
+		conn.Close()
+		return
+	}
+
+	// Validate token
+	if s.validateToken != nil && !s.validateToken(reg.Token) {
+		log.Warn().Str("agent_id", reg.AgentID).Msg("Agent registration rejected: invalid token")
+		s.sendMessage(conn, Message{
+			Type:      MsgTypeRegistered,
+			Timestamp: time.Now(),
+			Payload:   RegisteredPayload{Success: false, Message: "Invalid token"},
+		})
+		conn.Close()
+		return
+	}
+
+	// Create agent connection
+	ac := &agentConn{
+		conn: conn,
+		agent: ConnectedAgent{
+			AgentID:     reg.AgentID,
+			Hostname:    reg.Hostname,
+			Version:     reg.Version,
+			Platform:    reg.Platform,
+			Tags:        reg.Tags,
+			ConnectedAt: time.Now(),
+		},
+		done: make(chan struct{}),
+	}
+
+	// Register agent
+	s.mu.Lock()
+	// Close existing connection if any
+	if existing, ok := s.agents[reg.AgentID]; ok {
+		close(existing.done)
+		existing.conn.Close()
+	}
+	s.agents[reg.AgentID] = ac
+	s.mu.Unlock()
+
+	log.Info().
+		Str("agent_id", reg.AgentID).
+		Str("hostname", reg.Hostname).
+		Str("version", reg.Version).
+		Str("platform", reg.Platform).
+		Msg("Agent connected")
+
+	// Send registration success
+	s.sendMessage(conn, Message{
+		Type:      MsgTypeRegistered,
+		Timestamp: time.Now(),
+		Payload:   RegisteredPayload{Success: true, Message: "Registered"},
+	})
+
+	// Clear deadline for normal operation
+	conn.SetReadDeadline(time.Time{})
+
+	// Run read loop (blocking) - don't use goroutine, or HTTP handler will close connection
+	s.readLoop(ac)
+}
+
+func (s *Server) readLoop(ac *agentConn) {
+	defer func() {
+		s.mu.Lock()
+		if existing, ok := s.agents[ac.agent.AgentID]; ok && existing == ac {
+			delete(s.agents, ac.agent.AgentID)
+		}
+		s.mu.Unlock()
+		ac.conn.Close()
+		log.Info().Str("agent_id", ac.agent.AgentID).Msg("Agent disconnected")
+	}()
+
+	log.Debug().Str("agent_id", ac.agent.AgentID).Msg("Starting read loop for agent")
+
+	for {
+		select {
+		case <-ac.done:
+			log.Debug().Str("agent_id", ac.agent.AgentID).Msg("Read loop exiting: done channel closed")
+			return
+		default:
+		}
+
+		_, msgBytes, err := ac.conn.ReadMessage()
+		if err != nil {
+			log.Debug().Err(err).Str("agent_id", ac.agent.AgentID).Msg("Read loop exiting: read error")
+			if websocket.IsUnexpectedCloseError(err, websocket.CloseGoingAway, websocket.CloseAbnormalClosure) {
+				log.Error().Err(err).Str("agent_id", ac.agent.AgentID).Msg("WebSocket read error")
+			}
+			return
+		}
+
+		var msg Message
+		if err := json.Unmarshal(msgBytes, &msg); err != nil {
+			log.Error().Err(err).Str("agent_id", ac.agent.AgentID).Msg("Failed to parse message")
+			continue
+		}
+
+		switch msg.Type {
+		case MsgTypeAgentPing:
+			s.sendMessage(ac.conn, Message{
+				Type:      MsgTypePong,
+				Timestamp: time.Now(),
+			})
+
+		case MsgTypeCommandResult:
+			payloadBytes, _ := json.Marshal(msg.Payload)
+			var result CommandResultPayload
+			if err := json.Unmarshal(payloadBytes, &result); err != nil {
+				log.Error().Err(err).Msg("Failed to parse command result")
+				continue
+			}
+
+			s.mu.RLock()
+			ch, ok := s.pendingReqs[result.RequestID]
+			s.mu.RUnlock()
+
+			if ok {
+				select {
+				case ch <- result:
+				default:
+					log.Warn().Str("request_id", result.RequestID).Msg("Result channel full, dropping")
+				}
+			} else {
+				log.Warn().Str("request_id", result.RequestID).Msg("No pending request for result")
+			}
+		}
+	}
+}
+
+func (s *Server) sendMessage(conn *websocket.Conn, msg Message) error {
+	msgBytes, err := json.Marshal(msg)
+	if err != nil {
+		return err
+	}
+	return conn.WriteMessage(websocket.TextMessage, msgBytes)
+}
+
+// ExecuteCommand sends a command to an agent and waits for the result
+func (s *Server) ExecuteCommand(ctx context.Context, agentID string, cmd ExecuteCommandPayload) (*CommandResultPayload, error) {
+	s.mu.RLock()
+	ac, ok := s.agents[agentID]
+	s.mu.RUnlock()
+
+	if !ok {
+		return nil, fmt.Errorf("agent %s not connected", agentID)
+	}
+
+	// Create response channel
+	respCh := make(chan CommandResultPayload, 1)
+	s.mu.Lock()
+	s.pendingReqs[cmd.RequestID] = respCh
+	s.mu.Unlock()
+
+	defer func() {
+		s.mu.Lock()
+		delete(s.pendingReqs, cmd.RequestID)
+		s.mu.Unlock()
+	}()
+
+	// Send command
+	msg := Message{
+		Type:      MsgTypeExecuteCmd,
+		ID:        cmd.RequestID,
+		Timestamp: time.Now(),
+		Payload:   cmd,
+	}
+
+	ac.writeMu.Lock()
+	err := s.sendMessage(ac.conn, msg)
+	ac.writeMu.Unlock()
+
+	if err != nil {
+		return nil, fmt.Errorf("failed to send command: %w", err)
+	}
+
+	// Wait for result
+	timeout := time.Duration(cmd.Timeout) * time.Second
+	if timeout <= 0 {
+		timeout = 60 * time.Second
+	}
+
+	select {
+	case result := <-respCh:
+		return &result, nil
+	case <-time.After(timeout):
+		return nil, fmt.Errorf("command timed out after %v", timeout)
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
+}
+
+// ReadFile reads a file from an agent
+func (s *Server) ReadFile(ctx context.Context, agentID string, req ReadFilePayload) (*CommandResultPayload, error) {
+	s.mu.RLock()
+	ac, ok := s.agents[agentID]
+	s.mu.RUnlock()
+
+	if !ok {
+		return nil, fmt.Errorf("agent %s not connected", agentID)
+	}
+
+	// Create response channel
+	respCh := make(chan CommandResultPayload, 1)
+	s.mu.Lock()
+	s.pendingReqs[req.RequestID] = respCh
+	s.mu.Unlock()
+
+	defer func() {
+		s.mu.Lock()
+		delete(s.pendingReqs, req.RequestID)
+		s.mu.Unlock()
+	}()
+
+	// Send request
+	msg := Message{
+		Type:      MsgTypeReadFile,
+		ID:        req.RequestID,
+		Timestamp: time.Now(),
+		Payload:   req,
+	}
+
+	ac.writeMu.Lock()
+	err := s.sendMessage(ac.conn, msg)
+	ac.writeMu.Unlock()
+
+	if err != nil {
+		return nil, fmt.Errorf("failed to send read_file request: %w", err)
+	}
+
+	// Wait for result
+	timeout := 30 * time.Second
+	select {
+	case result := <-respCh:
+		return &result, nil
+	case <-time.After(timeout):
+		return nil, fmt.Errorf("read_file timed out after %v", timeout)
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	}
+}
+
+// GetConnectedAgents returns a list of currently connected agents
+func (s *Server) GetConnectedAgents() []ConnectedAgent {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	agents := make([]ConnectedAgent, 0, len(s.agents))
+	for _, ac := range s.agents {
+		agents = append(agents, ac.agent)
+	}
+	return agents
+}
+
+// IsAgentConnected checks if an agent is currently connected
+func (s *Server) IsAgentConnected(agentID string) bool {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+	_, ok := s.agents[agentID]
+	return ok
+}
+
+// GetAgentForHost finds the agent for a given hostname
+func (s *Server) GetAgentForHost(hostname string) (string, bool) {
+	s.mu.RLock()
+	defer s.mu.RUnlock()
+
+	for _, ac := range s.agents {
+		if ac.agent.Hostname == hostname {
+			return ac.agent.AgentID, true
+		}
+	}
+	return "", false
+}
--- a/internal/agentexec/types.go
+++ b/internal/agentexec/types.go
@@ -0,0 +1,84 @@
+package agentexec
+
+import (
+	"time"
+)
+
+// MessageType identifies the type of WebSocket message
+type MessageType string
+
+const (
+	// Agent -> Server messages
+	MsgTypeAgentRegister MessageType = "agent_register"
+	MsgTypeAgentPing     MessageType = "agent_ping"
+	MsgTypeCommandResult MessageType = "command_result"
+
+	// Server -> Agent messages
+	MsgTypeRegistered    MessageType = "registered"
+	MsgTypePong          MessageType = "pong"
+	MsgTypeExecuteCmd    MessageType = "execute_command"
+	MsgTypeReadFile      MessageType = "read_file"
+)
+
+// Message is the envelope for all WebSocket messages
+type Message struct {
+	Type      MessageType `json:"type"`
+	ID        string      `json:"id,omitempty"`         // Unique message ID for request/response correlation
+	Timestamp time.Time   `json:"timestamp"`
+	Payload   interface{} `json:"payload,omitempty"`
+}
+
+// AgentRegisterPayload is sent by agent on connection
+type AgentRegisterPayload struct {
+	AgentID   string   `json:"agent_id"`
+	Hostname  string   `json:"hostname"`
+	Version   string   `json:"version"`
+	Platform  string   `json:"platform"`  // "linux", "windows", "darwin"
+	Tags      []string `json:"tags,omitempty"`
+	Token     string   `json:"token"`     // API token for authentication
+}
+
+// RegisteredPayload is sent by server after successful registration
+type RegisteredPayload struct {
+	Success bool   `json:"success"`
+	Message string `json:"message,omitempty"`
+}
+
+// ExecuteCommandPayload is sent by server to request command execution
+type ExecuteCommandPayload struct {
+	RequestID  string `json:"request_id"`
+	Command    string `json:"command"`
+	TargetType string `json:"target_type"` // "host", "container", "vm"
+	TargetID   string `json:"target_id,omitempty"` // VMID for container/VM
+	Timeout    int    `json:"timeout,omitempty"` // seconds, 0 = default
+}
+
+// ReadFilePayload is sent by server to request file content
+type ReadFilePayload struct {
+	RequestID  string `json:"request_id"`
+	Path       string `json:"path"`
+	TargetType string `json:"target_type"`
+	TargetID   string `json:"target_id,omitempty"`
+	MaxBytes   int64  `json:"max_bytes,omitempty"` // 0 = default (1MB)
+}
+
+// CommandResultPayload is sent by agent with command execution result
+type CommandResultPayload struct {
+	RequestID string `json:"request_id"`
+	Success   bool   `json:"success"`
+	Stdout    string `json:"stdout,omitempty"`
+	Stderr    string `json:"stderr,omitempty"`
+	ExitCode  int    `json:"exit_code"`
+	Error     string `json:"error,omitempty"`
+	Duration  int64  `json:"duration_ms"`
+}
+
+// ConnectedAgent represents an agent connected via WebSocket
+type ConnectedAgent struct {
+	AgentID   string
+	Hostname  string
+	Version   string
+	Platform  string
+	Tags      []string
+	ConnectedAt time.Time
+}
--- a/internal/ai/config.go
+++ b/internal/ai/config.go
@@ -0,0 +1,19 @@
+// Package ai provides AI-powered infrastructure investigation and remediation.
+package ai
+
+import "github.com/rcourtman/pulse-go-rewrite/internal/config"
+
+// Re-export config types for convenience
+type Config = config.AIConfig
+
+// Provider constants (re-exported for convenience)
+const (
+	ProviderAnthropic = config.AIProviderAnthropic
+	ProviderOpenAI    = config.AIProviderOpenAI
+	ProviderOllama    = config.AIProviderOllama
+)
+
+// NewDefaultConfig returns a new AI config with sensible defaults
+func NewDefaultConfig() *Config {
+	return config.NewDefaultAIConfig()
+}
--- a/internal/ai/providers/anthropic.go
+++ b/internal/ai/providers/anthropic.go
@@ -0,0 +1,267 @@
+package providers
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"time"
+)
+
+const (
+	anthropicAPIURL     = "https://api.anthropic.com/v1/messages"
+	anthropicAPIVersion = "2023-06-01"
+)
+
+// AnthropicClient implements the Provider interface for Anthropic's Claude API
+type AnthropicClient struct {
+	apiKey string
+	model  string
+	client *http.Client
+}
+
+// NewAnthropicClient creates a new Anthropic API client
+func NewAnthropicClient(apiKey, model string) *AnthropicClient {
+	return &AnthropicClient{
+		apiKey: apiKey,
+		model:  model,
+		client: &http.Client{
+			Timeout: 120 * time.Second, // LLM responses can take a while
+		},
+	}
+}
+
+// Name returns the provider name
+func (c *AnthropicClient) Name() string {
+	return "anthropic"
+}
+
+// anthropicRequest is the request body for the Anthropic API
+type anthropicRequest struct {
+	Model       string             `json:"model"`
+	Messages    []anthropicMessage `json:"messages"`
+	MaxTokens   int                `json:"max_tokens"`
+	System      string             `json:"system,omitempty"`
+	Temperature float64            `json:"temperature,omitempty"`
+	Tools       []anthropicTool    `json:"tools,omitempty"`
+}
+
+type anthropicMessage struct {
+	Role    string        `json:"role"`
+	Content interface{}   `json:"content"` // Can be string or []anthropicContent
+}
+
+type anthropicTool struct {
+	Name        string                 `json:"name"`
+	Description string                 `json:"description"`
+	InputSchema map[string]interface{} `json:"input_schema"`
+}
+
+// anthropicResponse is the response from the Anthropic API
+type anthropicResponse struct {
+	ID           string `json:"id"`
+	Type         string `json:"type"`
+	Role         string `json:"role"`
+	Content      []anthropicContent `json:"content"`
+	Model        string `json:"model"`
+	StopReason   string `json:"stop_reason"`
+	StopSequence string `json:"stop_sequence,omitempty"`
+	Usage        anthropicUsage `json:"usage"`
+}
+
+type anthropicContent struct {
+	Type  string                 `json:"type"`            // "text" or "tool_use" or "tool_result"
+	Text  string                 `json:"text,omitempty"`
+	ID    string                 `json:"id,omitempty"`    // For tool_use
+	Name  string                 `json:"name,omitempty"`  // For tool_use
+	Input map[string]interface{} `json:"input,omitempty"` // For tool_use
+	ToolUseID string             `json:"tool_use_id,omitempty"` // For tool_result
+	Content   string             `json:"content,omitempty"`     // For tool_result (when it's a string)
+	IsError   bool               `json:"is_error,omitempty"`    // For tool_result
+}
+
+type anthropicUsage struct {
+	InputTokens  int `json:"input_tokens"`
+	OutputTokens int `json:"output_tokens"`
+}
+
+type anthropicError struct {
+	Type    string `json:"type"`
+	Error   anthropicErrorDetail `json:"error"`
+}
+
+type anthropicErrorDetail struct {
+	Type    string `json:"type"`
+	Message string `json:"message"`
+}
+
+// Chat sends a chat request to the Anthropic API
+func (c *AnthropicClient) Chat(ctx context.Context, req ChatRequest) (*ChatResponse, error) {
+	// Convert messages to Anthropic format
+	messages := make([]anthropicMessage, 0, len(req.Messages))
+	for _, m := range req.Messages {
+		// Anthropic doesn't use "system" role in messages array
+		if m.Role == "system" {
+			continue
+		}
+
+		// Handle tool results specially
+		if m.ToolResult != nil {
+			// Tool result message
+			messages = append(messages, anthropicMessage{
+				Role: "user",
+				Content: []anthropicContent{
+					{
+						Type:      "tool_result",
+						ToolUseID: m.ToolResult.ToolUseID,
+						Content:   m.ToolResult.Content,
+						IsError:   m.ToolResult.IsError,
+					},
+				},
+			})
+			continue
+		}
+
+		// Handle assistant messages with tool calls
+		if m.Role == "assistant" && len(m.ToolCalls) > 0 {
+			contentBlocks := make([]anthropicContent, 0)
+			if m.Content != "" {
+				contentBlocks = append(contentBlocks, anthropicContent{
+					Type: "text",
+					Text: m.Content,
+				})
+			}
+			for _, tc := range m.ToolCalls {
+				contentBlocks = append(contentBlocks, anthropicContent{
+					Type:  "tool_use",
+					ID:    tc.ID,
+					Name:  tc.Name,
+					Input: tc.Input,
+				})
+			}
+			messages = append(messages, anthropicMessage{
+				Role:    "assistant",
+				Content: contentBlocks,
+			})
+			continue
+		}
+
+		// Simple text message
+		messages = append(messages, anthropicMessage{
+			Role:    m.Role,
+			Content: m.Content,
+		})
+	}
+
+	// Use provided model or fall back to client default
+	model := req.Model
+	if model == "" {
+		model = c.model
+	}
+
+	// Set max tokens (Anthropic requires this)
+	maxTokens := req.MaxTokens
+	if maxTokens == 0 {
+		maxTokens = 4096
+	}
+
+	anthropicReq := anthropicRequest{
+		Model:     model,
+		Messages:  messages,
+		MaxTokens: maxTokens,
+		System:    req.System,
+	}
+
+	if req.Temperature > 0 {
+		anthropicReq.Temperature = req.Temperature
+	}
+
+	// Add tools if provided
+	if len(req.Tools) > 0 {
+		anthropicReq.Tools = make([]anthropicTool, len(req.Tools))
+		for i, t := range req.Tools {
+			anthropicReq.Tools[i] = anthropicTool{
+				Name:        t.Name,
+				Description: t.Description,
+				InputSchema: t.InputSchema,
+			}
+		}
+	}
+
+	body, err := json.Marshal(anthropicReq)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", anthropicAPIURL, bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	httpReq.Header.Set("Content-Type", "application/json")
+	httpReq.Header.Set("x-api-key", c.apiKey)
+	httpReq.Header.Set("anthropic-version", anthropicAPIVersion)
+
+	resp, err := c.client.Do(httpReq)
+	if err != nil {
+		return nil, fmt.Errorf("request failed: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		var errResp anthropicError
+		if err := json.Unmarshal(respBody, &errResp); err == nil && errResp.Error.Message != "" {
+			return nil, fmt.Errorf("API error (%d): %s", resp.StatusCode, errResp.Error.Message)
+		}
+		return nil, fmt.Errorf("API error (%d): %s", resp.StatusCode, string(respBody))
+	}
+
+	var anthropicResp anthropicResponse
+	if err := json.Unmarshal(respBody, &anthropicResp); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	// Extract content and tool calls from response
+	var textContent string
+	var toolCalls []ToolCall
+	for _, c := range anthropicResp.Content {
+		switch c.Type {
+		case "text":
+			textContent += c.Text
+		case "tool_use":
+			toolCalls = append(toolCalls, ToolCall{
+				ID:    c.ID,
+				Name:  c.Name,
+				Input: c.Input,
+			})
+		}
+	}
+
+	return &ChatResponse{
+		Content:      textContent,
+		Model:        anthropicResp.Model,
+		StopReason:   anthropicResp.StopReason,
+		ToolCalls:    toolCalls,
+		InputTokens:  anthropicResp.Usage.InputTokens,
+		OutputTokens: anthropicResp.Usage.OutputTokens,
+	}, nil
+}
+
+// TestConnection validates the API key by making a minimal request
+func (c *AnthropicClient) TestConnection(ctx context.Context) error {
+	// Make a minimal request to validate the API key
+	_, err := c.Chat(ctx, ChatRequest{
+		Messages: []Message{
+			{Role: "user", Content: "Hi"},
+		},
+		MaxTokens: 10, // Minimal tokens to save cost
+	})
+	return err
+}
--- a/internal/ai/providers/factory.go
+++ b/internal/ai/providers/factory.go
@@ -0,0 +1,38 @@
+package providers
+
+import (
+	"fmt"
+
+	"github.com/rcourtman/pulse-go-rewrite/internal/config"
+)
+
+// NewFromConfig creates a Provider based on the AIConfig settings
+func NewFromConfig(cfg *config.AIConfig) (Provider, error) {
+	if cfg == nil {
+		return nil, fmt.Errorf("AI config is nil")
+	}
+
+	if !cfg.Enabled {
+		return nil, fmt.Errorf("AI is not enabled")
+	}
+
+	switch cfg.Provider {
+	case config.AIProviderAnthropic:
+		if cfg.APIKey == "" {
+			return nil, fmt.Errorf("Anthropic API key is required")
+		}
+		return NewAnthropicClient(cfg.APIKey, cfg.GetModel()), nil
+
+	case config.AIProviderOpenAI:
+		if cfg.APIKey == "" {
+			return nil, fmt.Errorf("OpenAI API key is required")
+		}
+		return NewOpenAIClient(cfg.APIKey, cfg.GetModel(), cfg.GetBaseURL()), nil
+
+	case config.AIProviderOllama:
+		return NewOllamaClient(cfg.GetModel(), cfg.GetBaseURL()), nil
+
+	default:
+		return nil, fmt.Errorf("unknown provider: %s", cfg.Provider)
+	}
+}
--- a/internal/ai/providers/ollama.go
+++ b/internal/ai/providers/ollama.go
@@ -0,0 +1,173 @@
+package providers
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"time"
+)
+
+// OllamaClient implements the Provider interface for Ollama's local API
+type OllamaClient struct {
+	model   string
+	baseURL string
+	client  *http.Client
+}
+
+// NewOllamaClient creates a new Ollama API client
+func NewOllamaClient(model, baseURL string) *OllamaClient {
+	if baseURL == "" {
+		baseURL = "http://localhost:11434"
+	}
+	return &OllamaClient{
+		model:   model,
+		baseURL: baseURL,
+		client: &http.Client{
+			Timeout: 300 * time.Second, // Local models can be slow
+		},
+	}
+}
+
+// Name returns the provider name
+func (c *OllamaClient) Name() string {
+	return "ollama"
+}
+
+// ollamaRequest is the request body for the Ollama API
+type ollamaRequest struct {
+	Model    string          `json:"model"`
+	Messages []ollamaMessage `json:"messages"`
+	Stream   bool            `json:"stream"`
+	Options  *ollamaOptions  `json:"options,omitempty"`
+}
+
+type ollamaMessage struct {
+	Role    string `json:"role"`
+	Content string `json:"content"`
+}
+
+type ollamaOptions struct {
+	NumPredict  int     `json:"num_predict,omitempty"`
+	Temperature float64 `json:"temperature,omitempty"`
+}
+
+// ollamaResponse is the response from the Ollama API
+type ollamaResponse struct {
+	Model     string        `json:"model"`
+	CreatedAt string        `json:"created_at"`
+	Message   ollamaMessage `json:"message"`
+	Done      bool          `json:"done"`
+	DoneReason string       `json:"done_reason,omitempty"`
+	TotalDuration   int64  `json:"total_duration,omitempty"`
+	LoadDuration    int64  `json:"load_duration,omitempty"`
+	PromptEvalCount int    `json:"prompt_eval_count,omitempty"`
+	EvalCount       int    `json:"eval_count,omitempty"`
+}
+
+// Chat sends a chat request to the Ollama API
+func (c *OllamaClient) Chat(ctx context.Context, req ChatRequest) (*ChatResponse, error) {
+	// Convert messages to Ollama format
+	messages := make([]ollamaMessage, 0, len(req.Messages)+1)
+
+	// Add system message if provided
+	if req.System != "" {
+		messages = append(messages, ollamaMessage{
+			Role:    "system",
+			Content: req.System,
+		})
+	}
+
+	for _, m := range req.Messages {
+		messages = append(messages, ollamaMessage{
+			Role:    m.Role,
+			Content: m.Content,
+		})
+	}
+
+	// Use provided model or fall back to client default
+	model := req.Model
+	if model == "" {
+		model = c.model
+	}
+
+	ollamaReq := ollamaRequest{
+		Model:    model,
+		Messages: messages,
+		Stream:   false, // Non-streaming for now
+	}
+
+	if req.MaxTokens > 0 || req.Temperature > 0 {
+		ollamaReq.Options = &ollamaOptions{}
+		if req.MaxTokens > 0 {
+			ollamaReq.Options.NumPredict = req.MaxTokens
+		}
+		if req.Temperature > 0 {
+			ollamaReq.Options.Temperature = req.Temperature
+		}
+	}
+
+	body, err := json.Marshal(ollamaReq)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
+
+	url := c.baseURL + "/api/chat"
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", url, bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	httpReq.Header.Set("Content-Type", "application/json")
+
+	resp, err := c.client.Do(httpReq)
+	if err != nil {
+		return nil, fmt.Errorf("request failed: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		return nil, fmt.Errorf("API error (%d): %s", resp.StatusCode, string(respBody))
+	}
+
+	var ollamaResp ollamaResponse
+	if err := json.Unmarshal(respBody, &ollamaResp); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	return &ChatResponse{
+		Content:      ollamaResp.Message.Content,
+		Model:        ollamaResp.Model,
+		StopReason:   ollamaResp.DoneReason,
+		InputTokens:  ollamaResp.PromptEvalCount,
+		OutputTokens: ollamaResp.EvalCount,
+	}, nil
+}
+
+// TestConnection validates connectivity by checking the Ollama version endpoint
+func (c *OllamaClient) TestConnection(ctx context.Context) error {
+	url := c.baseURL + "/api/version"
+	httpReq, err := http.NewRequestWithContext(ctx, "GET", url, nil)
+	if err != nil {
+		return fmt.Errorf("failed to create request: %w", err)
+	}
+
+	resp, err := c.client.Do(httpReq)
+	if err != nil {
+		return fmt.Errorf("failed to connect to Ollama at %s: %w", c.baseURL, err)
+	}
+	defer resp.Body.Close()
+
+	if resp.StatusCode != http.StatusOK {
+		return fmt.Errorf("Ollama returned status %d", resp.StatusCode)
+	}
+
+	return nil
+}
--- a/internal/ai/providers/openai.go
+++ b/internal/ai/providers/openai.go
@@ -0,0 +1,188 @@
+package providers
+
+import (
+	"bytes"
+	"context"
+	"encoding/json"
+	"fmt"
+	"io"
+	"net/http"
+	"time"
+)
+
+const (
+	openaiAPIURL = "https://api.openai.com/v1/chat/completions"
+)
+
+// OpenAIClient implements the Provider interface for OpenAI's API
+type OpenAIClient struct {
+	apiKey  string
+	model   string
+	baseURL string
+	client  *http.Client
+}
+
+// NewOpenAIClient creates a new OpenAI API client
+func NewOpenAIClient(apiKey, model, baseURL string) *OpenAIClient {
+	if baseURL == "" {
+		baseURL = openaiAPIURL
+	}
+	return &OpenAIClient{
+		apiKey:  apiKey,
+		model:   model,
+		baseURL: baseURL,
+		client: &http.Client{
+			Timeout: 120 * time.Second,
+		},
+	}
+}
+
+// Name returns the provider name
+func (c *OpenAIClient) Name() string {
+	return "openai"
+}
+
+// openaiRequest is the request body for the OpenAI API
+type openaiRequest struct {
+	Model       string          `json:"model"`
+	Messages    []openaiMessage `json:"messages"`
+	MaxTokens   int             `json:"max_tokens,omitempty"`
+	Temperature float64         `json:"temperature,omitempty"`
+}
+
+type openaiMessage struct {
+	Role    string `json:"role"`
+	Content string `json:"content"`
+}
+
+// openaiResponse is the response from the OpenAI API
+type openaiResponse struct {
+	ID      string         `json:"id"`
+	Object  string         `json:"object"`
+	Created int64          `json:"created"`
+	Model   string         `json:"model"`
+	Choices []openaiChoice `json:"choices"`
+	Usage   openaiUsage    `json:"usage"`
+}
+
+type openaiChoice struct {
+	Index        int           `json:"index"`
+	Message      openaiMessage `json:"message"`
+	FinishReason string        `json:"finish_reason"`
+}
+
+type openaiUsage struct {
+	PromptTokens     int `json:"prompt_tokens"`
+	CompletionTokens int `json:"completion_tokens"`
+	TotalTokens      int `json:"total_tokens"`
+}
+
+type openaiError struct {
+	Error openaiErrorDetail `json:"error"`
+}
+
+type openaiErrorDetail struct {
+	Message string `json:"message"`
+	Type    string `json:"type"`
+	Code    string `json:"code"`
+}
+
+// Chat sends a chat request to the OpenAI API
+func (c *OpenAIClient) Chat(ctx context.Context, req ChatRequest) (*ChatResponse, error) {
+	// Convert messages to OpenAI format
+	messages := make([]openaiMessage, 0, len(req.Messages)+1)
+
+	// Add system message if provided
+	if req.System != "" {
+		messages = append(messages, openaiMessage{
+			Role:    "system",
+			Content: req.System,
+		})
+	}
+
+	for _, m := range req.Messages {
+		messages = append(messages, openaiMessage{
+			Role:    m.Role,
+			Content: m.Content,
+		})
+	}
+
+	// Use provided model or fall back to client default
+	model := req.Model
+	if model == "" {
+		model = c.model
+	}
+
+	openaiReq := openaiRequest{
+		Model:    model,
+		Messages: messages,
+	}
+
+	if req.MaxTokens > 0 {
+		openaiReq.MaxTokens = req.MaxTokens
+	}
+
+	if req.Temperature > 0 {
+		openaiReq.Temperature = req.Temperature
+	}
+
+	body, err := json.Marshal(openaiReq)
+	if err != nil {
+		return nil, fmt.Errorf("failed to marshal request: %w", err)
+	}
+
+	httpReq, err := http.NewRequestWithContext(ctx, "POST", c.baseURL, bytes.NewReader(body))
+	if err != nil {
+		return nil, fmt.Errorf("failed to create request: %w", err)
+	}
+
+	httpReq.Header.Set("Content-Type", "application/json")
+	httpReq.Header.Set("Authorization", "Bearer "+c.apiKey)
+
+	resp, err := c.client.Do(httpReq)
+	if err != nil {
+		return nil, fmt.Errorf("request failed: %w", err)
+	}
+	defer resp.Body.Close()
+
+	respBody, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("failed to read response: %w", err)
+	}
+
+	if resp.StatusCode != http.StatusOK {
+		var errResp openaiError
+		if err := json.Unmarshal(respBody, &errResp); err == nil && errResp.Error.Message != "" {
+			return nil, fmt.Errorf("API error (%d): %s", resp.StatusCode, errResp.Error.Message)
+		}
+		return nil, fmt.Errorf("API error (%d): %s", resp.StatusCode, string(respBody))
+	}
+
+	var openaiResp openaiResponse
+	if err := json.Unmarshal(respBody, &openaiResp); err != nil {
+		return nil, fmt.Errorf("failed to parse response: %w", err)
+	}
+
+	if len(openaiResp.Choices) == 0 {
+		return nil, fmt.Errorf("no response choices returned")
+	}
+
+	return &ChatResponse{
+		Content:      openaiResp.Choices[0].Message.Content,
+		Model:        openaiResp.Model,
+		StopReason:   openaiResp.Choices[0].FinishReason,
+		InputTokens:  openaiResp.Usage.PromptTokens,
+		OutputTokens: openaiResp.Usage.CompletionTokens,
+	}, nil
+}
+
+// TestConnection validates the API key by making a minimal request
+func (c *OpenAIClient) TestConnection(ctx context.Context) error {
+	_, err := c.Chat(ctx, ChatRequest{
+		Messages: []Message{
+			{Role: "user", Content: "Hi"},
+		},
+		MaxTokens: 10,
+	})
+	return err
+}
--- a/internal/ai/providers/provider.go
+++ b/internal/ai/providers/provider.go
@@ -0,0 +1,67 @@
+// Package providers contains AI provider client implementations
+package providers
+
+import (
+	"context"
+)
+
+// Message represents a chat message
+type Message struct {
+	Role       string        `json:"role"`    // "user", "assistant", "system"
+	Content    string        `json:"content"` // Text content (simple case)
+	ToolCalls  []ToolCall    `json:"tool_calls,omitempty"` // For assistant messages with tool calls
+	ToolResult *ToolResult   `json:"tool_result,omitempty"` // For user messages with tool results
+}
+
+// ToolCall represents a tool invocation from the AI
+type ToolCall struct {
+	ID    string                 `json:"id"`
+	Name  string                 `json:"name"`
+	Input map[string]interface{} `json:"input"`
+}
+
+// ToolResult represents the result of a tool execution
+type ToolResult struct {
+	ToolUseID string `json:"tool_use_id"`
+	Content   string `json:"content"`
+	IsError   bool   `json:"is_error,omitempty"`
+}
+
+// Tool represents an AI tool definition
+type Tool struct {
+	Name        string                `json:"name"`
+	Description string                `json:"description"`
+	InputSchema map[string]interface{} `json:"input_schema"`
+}
+
+// ChatRequest represents a request to the AI provider
+type ChatRequest struct {
+	Messages    []Message `json:"messages"`
+	Model       string    `json:"model"`
+	MaxTokens   int       `json:"max_tokens,omitempty"`
+	Temperature float64   `json:"temperature,omitempty"`
+	System      string    `json:"system,omitempty"` // System prompt (Anthropic style)
+	Tools       []Tool    `json:"tools,omitempty"`  // Available tools
+}
+
+// ChatResponse represents a response from the AI provider
+type ChatResponse struct {
+	Content      string      `json:"content"`
+	Model        string      `json:"model"`
+	StopReason   string      `json:"stop_reason,omitempty"` // "end_turn", "tool_use"
+	ToolCalls    []ToolCall  `json:"tool_calls,omitempty"` // Tool invocations
+	InputTokens  int         `json:"input_tokens,omitempty"`
+	OutputTokens int         `json:"output_tokens,omitempty"`
+}
+
+// Provider defines the interface for AI providers
+type Provider interface {
+	// Chat sends a chat request and returns the response
+	Chat(ctx context.Context, req ChatRequest) (*ChatResponse, error)
+
+	// TestConnection validates the API key and connectivity
+	TestConnection(ctx context.Context) error
+
+	// Name returns the provider name
+	Name() string
+}
--- a/internal/ai/service.go
+++ b/internal/ai/service.go
--- a/internal/api/ai_handlers.go
+++ b/internal/api/ai_handlers.go
@@ -0,0 +1,614 @@
+package api
+
+import (
+	"context"
+	"encoding/json"
+	"net/http"
+	"strings"
+	"time"
+
+	"github.com/rcourtman/pulse-go-rewrite/internal/agentexec"
+	"github.com/rcourtman/pulse-go-rewrite/internal/ai"
+	"github.com/rcourtman/pulse-go-rewrite/internal/config"
+	"github.com/rcourtman/pulse-go-rewrite/internal/utils"
+	"github.com/rs/zerolog/log"
+)
+
+// AISettingsHandler handles AI settings endpoints
+type AISettingsHandler struct {
+	config      *config.Config
+	persistence *config.ConfigPersistence
+	aiService   *ai.Service
+	agentServer *agentexec.Server
+}
+
+// NewAISettingsHandler creates a new AI settings handler
+func NewAISettingsHandler(cfg *config.Config, persistence *config.ConfigPersistence, agentServer *agentexec.Server) *AISettingsHandler {
+	aiService := ai.NewService(persistence, agentServer)
+	if err := aiService.LoadConfig(); err != nil {
+		log.Warn().Err(err).Msg("Failed to load AI config on startup")
+	}
+
+	return &AISettingsHandler{
+		config:      cfg,
+		persistence: persistence,
+		aiService:   aiService,
+		agentServer: agentServer,
+	}
+}
+
+// SetConfig updates the configuration reference used by the handler.
+func (h *AISettingsHandler) SetConfig(cfg *config.Config) {
+	if cfg == nil {
+		return
+	}
+	h.config = cfg
+}
+
+// SetStateProvider sets the state provider for infrastructure context
+func (h *AISettingsHandler) SetStateProvider(sp ai.StateProvider) {
+	h.aiService.SetStateProvider(sp)
+}
+
+// AISettingsResponse is returned by GET /api/settings/ai
+// API key is masked for security
+type AISettingsResponse struct {
+	Enabled        bool   `json:"enabled"`
+	Provider       string `json:"provider"`
+	APIKeySet      bool   `json:"api_key_set"` // true if API key is configured (never expose actual key)
+	Model          string `json:"model"`
+	BaseURL        string `json:"base_url,omitempty"`
+	Configured     bool   `json:"configured"`      // true if AI is ready to use
+	AutonomousMode bool   `json:"autonomous_mode"` // true if AI can execute without approval
+	CustomContext  string `json:"custom_context"`  // user-provided infrastructure context
+}
+
+// AISettingsUpdateRequest is the request body for PUT /api/settings/ai
+type AISettingsUpdateRequest struct {
+	Enabled        *bool   `json:"enabled,omitempty"`
+	Provider       *string `json:"provider,omitempty"`
+	APIKey         *string `json:"api_key,omitempty"` // empty string clears, null preserves
+	Model          *string `json:"model,omitempty"`
+	BaseURL        *string `json:"base_url,omitempty"`
+	AutonomousMode *bool   `json:"autonomous_mode,omitempty"`
+	CustomContext  *string `json:"custom_context,omitempty"` // user-provided infrastructure context
+}
+
+// HandleGetAISettings returns the current AI settings (GET /api/settings/ai)
+func (h *AISettingsHandler) HandleGetAISettings(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodGet {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	settings, err := h.persistence.LoadAIConfig()
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to load AI settings")
+		http.Error(w, "Failed to load AI settings", http.StatusInternalServerError)
+		return
+	}
+
+	if settings == nil {
+		settings = config.NewDefaultAIConfig()
+	}
+
+	response := AISettingsResponse{
+		Enabled:        settings.Enabled,
+		Provider:       settings.Provider,
+		APIKeySet:      settings.APIKey != "",
+		Model:          settings.GetModel(),
+		BaseURL:        settings.BaseURL,
+		Configured:     settings.IsConfigured(),
+		AutonomousMode: settings.AutonomousMode,
+		CustomContext:  settings.CustomContext,
+	}
+
+	if err := utils.WriteJSONResponse(w, response); err != nil {
+		log.Error().Err(err).Msg("Failed to write AI settings response")
+	}
+}
+
+// HandleUpdateAISettings updates AI settings (PUT /api/settings/ai)
+func (h *AISettingsHandler) HandleUpdateAISettings(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPut && r.Method != http.MethodPost {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	// Require admin authentication
+	if !CheckAuth(h.config, w, r) {
+		return
+	}
+
+	// Check proxy auth admin status if applicable
+	if h.config.ProxyAuthSecret != "" {
+		if valid, username, isAdmin := CheckProxyAuth(h.config, r); valid && !isAdmin {
+			log.Warn().
+				Str("ip", r.RemoteAddr).
+				Str("path", r.URL.Path).
+				Str("method", r.Method).
+				Str("username", username).
+				Msg("Non-admin user attempted to update AI settings")
+			w.Header().Set("Content-Type", "application/json")
+			w.WriteHeader(http.StatusForbidden)
+			_ = json.NewEncoder(w).Encode(map[string]string{"error": "Admin privileges required"})
+			return
+		}
+	}
+
+	// Load existing settings
+	settings, err := h.persistence.LoadAIConfig()
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to load existing AI settings")
+		settings = config.NewDefaultAIConfig()
+	}
+	if settings == nil {
+		settings = config.NewDefaultAIConfig()
+	}
+
+	// Parse request
+	r.Body = http.MaxBytesReader(w, r.Body, 16*1024)
+	var req AISettingsUpdateRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "Invalid request body", http.StatusBadRequest)
+		return
+	}
+
+	// Validate and apply updates
+	if req.Provider != nil {
+		provider := strings.ToLower(strings.TrimSpace(*req.Provider))
+		switch provider {
+		case config.AIProviderAnthropic, config.AIProviderOpenAI, config.AIProviderOllama:
+			settings.Provider = provider
+		default:
+			http.Error(w, "Invalid provider. Must be 'anthropic', 'openai', or 'ollama'", http.StatusBadRequest)
+			return
+		}
+	}
+
+	if req.APIKey != nil {
+		// Empty string clears the API key
+		settings.APIKey = strings.TrimSpace(*req.APIKey)
+	}
+
+	if req.Model != nil {
+		settings.Model = strings.TrimSpace(*req.Model)
+	}
+
+	if req.BaseURL != nil {
+		settings.BaseURL = strings.TrimSpace(*req.BaseURL)
+	}
+
+	if req.AutonomousMode != nil {
+		settings.AutonomousMode = *req.AutonomousMode
+	}
+
+	if req.CustomContext != nil {
+		settings.CustomContext = strings.TrimSpace(*req.CustomContext)
+	}
+
+	if req.Enabled != nil {
+		// Only allow enabling if properly configured
+		if *req.Enabled {
+			switch settings.Provider {
+			case config.AIProviderAnthropic, config.AIProviderOpenAI:
+				if settings.APIKey == "" {
+					http.Error(w, "Cannot enable AI: API key is required for "+settings.Provider, http.StatusBadRequest)
+					return
+				}
+			case config.AIProviderOllama:
+				// Ollama doesn't need API key, but needs base URL (or will use default)
+				if settings.BaseURL == "" {
+					settings.BaseURL = config.DefaultOllamaBaseURL
+				}
+			}
+		}
+		settings.Enabled = *req.Enabled
+	}
+
+	// Save settings
+	if err := h.persistence.SaveAIConfig(*settings); err != nil {
+		log.Error().Err(err).Msg("Failed to save AI settings")
+		http.Error(w, "Failed to save settings", http.StatusInternalServerError)
+		return
+	}
+
+	// Reload the AI service with new settings
+	if err := h.aiService.Reload(); err != nil {
+		log.Warn().Err(err).Msg("Failed to reload AI service after settings update")
+	}
+
+	log.Info().
+		Bool("enabled", settings.Enabled).
+		Str("provider", settings.Provider).
+		Str("model", settings.GetModel()).
+		Msg("AI settings updated")
+
+	// Return updated settings
+	response := AISettingsResponse{
+		Enabled:        settings.Enabled,
+		Provider:       settings.Provider,
+		APIKeySet:      settings.APIKey != "",
+		Model:          settings.GetModel(),
+		BaseURL:        settings.BaseURL,
+		Configured:     settings.IsConfigured(),
+		AutonomousMode: settings.AutonomousMode,
+		CustomContext:  settings.CustomContext,
+	}
+
+	if err := utils.WriteJSONResponse(w, response); err != nil {
+		log.Error().Err(err).Msg("Failed to write AI settings update response")
+	}
+}
+
+// HandleTestAIConnection tests the AI provider connection (POST /api/ai/test)
+func (h *AISettingsHandler) HandleTestAIConnection(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPost {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	// Require admin authentication
+	if !CheckAuth(h.config, w, r) {
+		return
+	}
+
+	ctx, cancel := context.WithTimeout(r.Context(), 30*time.Second)
+	defer cancel()
+
+	var testResult struct {
+		Success bool   `json:"success"`
+		Message string `json:"message"`
+		Model   string `json:"model,omitempty"`
+	}
+
+	err := h.aiService.TestConnection(ctx)
+	if err != nil {
+		testResult.Success = false
+		testResult.Message = err.Error()
+	} else {
+		cfg := h.aiService.GetConfig()
+		testResult.Success = true
+		testResult.Message = "Connection successful"
+		if cfg != nil {
+			testResult.Model = cfg.GetModel()
+		}
+	}
+
+	if err := utils.WriteJSONResponse(w, testResult); err != nil {
+		log.Error().Err(err).Msg("Failed to write AI test response")
+	}
+}
+
+// AIExecuteRequest is the request body for POST /api/ai/execute
+// AIConversationMessage represents a message in conversation history
+type AIConversationMessage struct {
+	Role    string `json:"role"`    // "user" or "assistant"
+	Content string `json:"content"`
+}
+
+type AIExecuteRequest struct {
+	Prompt     string                   `json:"prompt"`
+	TargetType string                   `json:"target_type,omitempty"` // "host", "container", "vm", "node"
+	TargetID   string                   `json:"target_id,omitempty"`
+	Context    map[string]interface{}   `json:"context,omitempty"` // Current metrics, state, etc.
+	History    []AIConversationMessage  `json:"history,omitempty"` // Previous conversation messages
+}
+
+// AIExecuteResponse is the response from POST /api/ai/execute
+type AIExecuteResponse struct {
+	Content      string                 `json:"content"`
+	Model        string                 `json:"model"`
+	InputTokens  int                    `json:"input_tokens"`
+	OutputTokens int                    `json:"output_tokens"`
+	ToolCalls    []ai.ToolExecution     `json:"tool_calls,omitempty"` // Commands that were executed
+}
+
+// HandleExecute executes an AI prompt (POST /api/ai/execute)
+func (h *AISettingsHandler) HandleExecute(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPost {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	// Require authentication
+	if !CheckAuth(h.config, w, r) {
+		return
+	}
+
+	// Check if AI is enabled
+	if !h.aiService.IsEnabled() {
+		http.Error(w, "AI is not enabled or configured", http.StatusBadRequest)
+		return
+	}
+
+	// Parse request
+	r.Body = http.MaxBytesReader(w, r.Body, 64*1024) // 64KB max
+	var req AIExecuteRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "Invalid request body", http.StatusBadRequest)
+		return
+	}
+
+	if strings.TrimSpace(req.Prompt) == "" {
+		http.Error(w, "Prompt is required", http.StatusBadRequest)
+		return
+	}
+
+	// Execute the prompt with a timeout
+	ctx, cancel := context.WithTimeout(r.Context(), 120*time.Second)
+	defer cancel()
+
+	resp, err := h.aiService.Execute(ctx, ai.ExecuteRequest{
+		Prompt:     req.Prompt,
+		TargetType: req.TargetType,
+		TargetID:   req.TargetID,
+		Context:    req.Context,
+	})
+	if err != nil {
+		log.Error().Err(err).Msg("AI execution failed")
+		http.Error(w, "AI request failed: "+err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	response := AIExecuteResponse{
+		Content:      resp.Content,
+		Model:        resp.Model,
+		InputTokens:  resp.InputTokens,
+		OutputTokens: resp.OutputTokens,
+		ToolCalls:    resp.ToolCalls,
+	}
+
+	if err := utils.WriteJSONResponse(w, response); err != nil {
+		log.Error().Err(err).Msg("Failed to write AI execute response")
+	}
+}
+
+// HandleExecuteStream executes an AI prompt with SSE streaming (POST /api/ai/execute/stream)
+func (h *AISettingsHandler) HandleExecuteStream(w http.ResponseWriter, r *http.Request) {
+	// Handle CORS for dev mode (frontend on different port)
+	origin := r.Header.Get("Origin")
+	if origin != "" {
+		// Must use specific origin (not *) when credentials are included
+		w.Header().Set("Access-Control-Allow-Origin", origin)
+		w.Header().Set("Access-Control-Allow-Credentials", "true")
+		w.Header().Set("Access-Control-Allow-Methods", "POST, OPTIONS")
+		w.Header().Set("Access-Control-Allow-Headers", "Content-Type, Accept, Cookie")
+		w.Header().Set("Vary", "Origin")
+	}
+
+	// Handle preflight
+	if r.Method == http.MethodOptions {
+		w.WriteHeader(http.StatusOK)
+		return
+	}
+
+	if r.Method != http.MethodPost {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	// Require authentication
+	if !CheckAuth(h.config, w, r) {
+		return
+	}
+
+	// Check if AI is enabled
+	if !h.aiService.IsEnabled() {
+		http.Error(w, "AI is not enabled or configured", http.StatusBadRequest)
+		return
+	}
+
+	// Parse request
+	r.Body = http.MaxBytesReader(w, r.Body, 64*1024) // 64KB max
+	var req AIExecuteRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "Invalid request body", http.StatusBadRequest)
+		return
+	}
+
+	if strings.TrimSpace(req.Prompt) == "" {
+		http.Error(w, "Prompt is required", http.StatusBadRequest)
+		return
+	}
+
+	log.Info().
+		Str("prompt", req.Prompt).
+		Str("target_type", req.TargetType).
+		Str("target_id", req.TargetID).
+		Msg("AI streaming request started")
+
+	// Set up SSE headers
+	// IMPORTANT: Set headers BEFORE any writes to prevent Go from auto-adding Transfer-Encoding: chunked
+	w.Header().Set("Content-Type", "text/event-stream")
+	w.Header().Set("Cache-Control", "no-cache")
+	w.Header().Set("Connection", "keep-alive")
+	w.Header().Set("X-Accel-Buffering", "no") // Disable nginx buffering
+	// Prevent chunked encoding which causes "Invalid character in chunk size" errors in Vite proxy
+	w.Header().Set("Transfer-Encoding", "identity")
+
+	flusher, ok := w.(http.Flusher)
+	if !ok {
+		http.Error(w, "Streaming not supported", http.StatusInternalServerError)
+		return
+	}
+
+	// Disable the server's write deadline for this SSE connection
+	// This is critical for long-running AI requests that can take several minutes
+	rc := http.NewResponseController(w)
+	if err := rc.SetWriteDeadline(time.Time{}); err != nil {
+		log.Warn().Err(err).Msg("Failed to disable write deadline for SSE")
+		// Continue anyway - heartbeats should help keep connection alive
+	}
+
+	// Flush headers immediately
+	flusher.Flush()
+
+	// Create context with timeout (5 minutes for complex analysis with multiple tool calls)
+	// Use background context to avoid browser disconnect canceling the request
+	ctx, cancel := context.WithTimeout(context.Background(), 300*time.Second)
+	defer cancel()
+
+	// Set up heartbeat to keep connection alive during long tool executions
+	// NOTE: We don't check r.Context().Done() because Vite proxy may close
+	// the request context prematurely. We detect real disconnection via write failures.
+	heartbeatDone := make(chan struct{})
+	go func() {
+		ticker := time.NewTicker(5 * time.Second)
+		defer ticker.Stop()
+		for {
+			select {
+			case <-ticker.C:
+				// Send SSE comment as heartbeat
+				_, err := w.Write([]byte(": heartbeat\n\n"))
+				if err != nil {
+					log.Debug().Err(err).Msg("Heartbeat write failed, client disconnected")
+					cancel() // Cancel the AI request
+					return
+				}
+				flusher.Flush()
+				log.Debug().Msg("Sent SSE heartbeat")
+			case <-heartbeatDone:
+				return
+			}
+		}
+	}()
+	defer close(heartbeatDone)
+
+	// Stream callback - write SSE events
+	callback := func(event ai.StreamEvent) {
+		data, err := json.Marshal(event)
+		if err != nil {
+			log.Error().Err(err).Msg("Failed to marshal stream event")
+			return
+		}
+
+		log.Debug().
+			Str("event_type", event.Type).
+			Msg("Streaming AI event")
+
+		// SSE format: data: <json>\n\n
+		_, writeErr := w.Write([]byte("data: " + string(data) + "\n\n"))
+		if writeErr != nil {
+			log.Debug().Err(writeErr).Msg("Failed to write SSE event (client may have disconnected)")
+			return
+		}
+		flusher.Flush()
+	}
+
+	// Convert history from API type to service type
+	var history []ai.ConversationMessage
+	for _, msg := range req.History {
+		history = append(history, ai.ConversationMessage{
+			Role:    msg.Role,
+			Content: msg.Content,
+		})
+	}
+
+	// Execute with streaming
+	resp, err := h.aiService.ExecuteStream(ctx, ai.ExecuteRequest{
+		Prompt:     req.Prompt,
+		TargetType: req.TargetType,
+		TargetID:   req.TargetID,
+		Context:    req.Context,
+		History:    history,
+	}, callback)
+
+	if err != nil {
+		log.Error().Err(err).Msg("AI streaming execution failed")
+		// Send error event
+		errEvent := ai.StreamEvent{Type: "error", Data: err.Error()}
+		data, _ := json.Marshal(errEvent)
+		_, _ = w.Write([]byte("data: " + string(data) + "\n\n"))
+		flusher.Flush()
+		return
+	}
+
+	log.Info().
+		Str("model", resp.Model).
+		Int("input_tokens", resp.InputTokens).
+		Int("output_tokens", resp.OutputTokens).
+		Int("tool_calls", len(resp.ToolCalls)).
+		Msg("AI streaming request completed")
+
+	// Send final response with metadata
+	finalEvent := struct {
+		Type         string             `json:"type"`
+		Model        string             `json:"model"`
+		InputTokens  int                `json:"input_tokens"`
+		OutputTokens int                `json:"output_tokens"`
+		ToolCalls    []ai.ToolExecution `json:"tool_calls,omitempty"`
+	}{
+		Type:         "complete",
+		Model:        resp.Model,
+		InputTokens:  resp.InputTokens,
+		OutputTokens: resp.OutputTokens,
+		ToolCalls:    resp.ToolCalls,
+	}
+	data, _ := json.Marshal(finalEvent)
+	_, _ = w.Write([]byte("data: " + string(data) + "\n\n"))
+	flusher.Flush()
+}
+
+// AIRunCommandRequest is the request body for POST /api/ai/run-command
+type AIRunCommandRequest struct {
+	Command    string `json:"command"`
+	TargetType string `json:"target_type"`
+	TargetID   string `json:"target_id"`
+	RunOnHost  bool   `json:"run_on_host"`
+	VMID       string `json:"vmid,omitempty"`
+}
+
+// HandleRunCommand executes a single approved command (POST /api/ai/run-command)
+func (h *AISettingsHandler) HandleRunCommand(w http.ResponseWriter, r *http.Request) {
+	if r.Method != http.MethodPost {
+		http.Error(w, "Method not allowed", http.StatusMethodNotAllowed)
+		return
+	}
+
+	// Require authentication
+	if !CheckAuth(h.config, w, r) {
+		return
+	}
+
+	// Parse request
+	r.Body = http.MaxBytesReader(w, r.Body, 16*1024)
+	var req AIRunCommandRequest
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		http.Error(w, "Invalid request body", http.StatusBadRequest)
+		return
+	}
+
+	if strings.TrimSpace(req.Command) == "" {
+		http.Error(w, "Command is required", http.StatusBadRequest)
+		return
+	}
+
+	log.Info().
+		Str("command", req.Command).
+		Str("target_type", req.TargetType).
+		Str("target_id", req.TargetID).
+		Bool("run_on_host", req.RunOnHost).
+		Msg("Executing approved command")
+
+	// Execute with timeout
+	ctx, cancel := context.WithTimeout(r.Context(), 120*time.Second)
+	defer cancel()
+
+	resp, err := h.aiService.RunCommand(ctx, ai.RunCommandRequest{
+		Command:    req.Command,
+		TargetType: req.TargetType,
+		TargetID:   req.TargetID,
+		RunOnHost:  req.RunOnHost,
+		VMID:       req.VMID,
+	})
+	if err != nil {
+		log.Error().Err(err).Msg("Failed to execute command")
+		http.Error(w, "Failed to execute command: "+err.Error(), http.StatusInternalServerError)
+		return
+	}
+
+	if err := utils.WriteJSONResponse(w, resp); err != nil {
+		log.Error().Err(err).Msg("Failed to write run command response")
+	}
+}
--- a/internal/api/router.go
+++ b/internal/api/router.go
@@ -26,6 +26,7 @@ import (
 	"time"

 	"github.com/rcourtman/pulse-go-rewrite/internal/agentbinaries"
+	"github.com/rcourtman/pulse-go-rewrite/internal/agentexec"
 	"github.com/rcourtman/pulse-go-rewrite/internal/auth"
 	"github.com/rcourtman/pulse-go-rewrite/internal/config"
 	"github.com/rcourtman/pulse-go-rewrite/internal/models"
@@ -51,6 +52,8 @@ type Router struct {
 	hostAgentHandlers         *HostAgentHandlers
 	temperatureProxyHandlers  *TemperatureProxyHandlers
 	systemSettingsHandler     *SystemSettingsHandler
+	aiSettingsHandler         *AISettingsHandler
+	agentExecServer           *agentexec.Server
 	wsHub                     *websocket.Hub
 	reloadFunc                func() error
 	updateManager             *updates.Manager
@@ -994,6 +997,39 @@ func (r *Router) setupRoutes() {
 	r.mux.HandleFunc("/api/system/proxy-public-key", r.handleProxyPublicKey)
 	// Old API token endpoints removed - now using /api/security/regenerate-token

+	// Agent execution server for AI tool use
+	r.agentExecServer = agentexec.NewServer(func(token string) bool {
+		// Validate agent tokens using the API tokens system
+		if r.config == nil {
+			return false
+		}
+		// First check the new API tokens system
+		if _, ok := r.config.ValidateAPIToken(token); ok {
+			return true
+		}
+		// Fall back to legacy single token if set
+		if r.config.APIToken != "" {
+			return auth.CompareAPIToken(token, r.config.APIToken)
+		}
+		return false
+	})
+
+	// AI settings endpoints
+	r.aiSettingsHandler = NewAISettingsHandler(r.config, r.persistence, r.agentExecServer)
+	// Inject state provider so AI has access to full infrastructure context (VMs, containers, IPs)
+	if r.monitor != nil {
+		r.aiSettingsHandler.SetStateProvider(r.monitor)
+	}
+	r.mux.HandleFunc("/api/settings/ai", RequireAdmin(r.config, RequireScope(config.ScopeSettingsRead, r.aiSettingsHandler.HandleGetAISettings)))
+	r.mux.HandleFunc("/api/settings/ai/update", RequireAdmin(r.config, RequireScope(config.ScopeSettingsWrite, r.aiSettingsHandler.HandleUpdateAISettings)))
+	r.mux.HandleFunc("/api/ai/test", RequireAdmin(r.config, RequireScope(config.ScopeSettingsWrite, r.aiSettingsHandler.HandleTestAIConnection)))
+	r.mux.HandleFunc("/api/ai/execute", RequireAuth(r.config, r.aiSettingsHandler.HandleExecute))
+	r.mux.HandleFunc("/api/ai/execute/stream", RequireAuth(r.config, r.aiSettingsHandler.HandleExecuteStream))
+	r.mux.HandleFunc("/api/ai/run-command", RequireAuth(r.config, r.aiSettingsHandler.HandleRunCommand))
+
+	// Agent WebSocket for AI command execution
+	r.mux.HandleFunc("/api/agent/ws", r.handleAgentWebSocket)
+
 	// Docker agent download endpoints
 	r.mux.HandleFunc("/install-docker-agent.sh", r.handleDownloadInstallScript) // Serves the Docker agent install script
 	r.mux.HandleFunc("/install-container-agent.sh", r.handleDownloadContainerAgentInstallScript)
@@ -1029,6 +1065,15 @@ func (r *Router) setupRoutes() {

 }

+// handleAgentWebSocket handles WebSocket connections from agents for AI command execution
+func (r *Router) handleAgentWebSocket(w http.ResponseWriter, req *http.Request) {
+	if r.agentExecServer == nil {
+		http.Error(w, "Agent execution not available", http.StatusServiceUnavailable)
+		return
+	}
+	r.agentExecServer.HandleWebSocket(w, req)
+}
+
 func (r *Router) handleVerifyTemperatureSSH(w http.ResponseWriter, req *http.Request) {
 	if r.configHandlers == nil {
 		http.Error(w, "Service unavailable", http.StatusServiceUnavailable)
@@ -1289,6 +1334,10 @@ func (r *Router) ServeHTTP(w http.ResponseWriter, req *http.Request) {
 			w.Header().Set("Access-Control-Allow-Methods", "GET, POST, PUT, DELETE, OPTIONS")
 			w.Header().Set("Access-Control-Allow-Headers", "Content-Type, Authorization, X-API-Token, X-CSRF-Token, X-Setup-Token")
 			w.Header().Set("Access-Control-Expose-Headers", "X-CSRF-Token, X-Authenticated-User, X-Auth-Method")
+			// Allow credentials when origin is specific (not *)
+			if r.config.AllowedOrigins != "*" {
+				w.Header().Set("Access-Control-Allow-Credentials", "true")
+			}
 		}

 		// Handle preflight requests
@@ -1351,6 +1400,7 @@ func (r *Router) ServeHTTP(w http.ResponseWriter, req *http.Request) {
 				"/install.ps1",                                       // Unified agent Windows installer
 				"/download/pulse-agent",                              // Unified agent binary
 				"/api/agent/version",                                 // Agent update checks need to work before auth
+				"/api/agent/ws",                                      // Agent WebSocket has its own auth via registration
 				"/api/server/info",                                   // Server info for installer script
 				"/api/install/install-sensor-proxy.sh",               // Temperature proxy installer fallback
 				"/api/install/pulse-sensor-proxy",                    // Temperature proxy binary fallback
--- a/internal/config/ai.go
+++ b/internal/config/ai.go
@@ -0,0 +1,82 @@
+package config
+
+// AIConfig holds AI feature configuration
+// This is stored in ai.enc (encrypted) in the config directory
+type AIConfig struct {
+	Enabled        bool   `json:"enabled"`
+	Provider       string `json:"provider"`        // "anthropic", "openai", "ollama"
+	APIKey         string `json:"api_key"`         // encrypted at rest (not needed for ollama)
+	Model          string `json:"model"`           // e.g., "claude-opus-4-5-20250514", "gpt-4o", "llama3"
+	BaseURL        string `json:"base_url"`        // custom endpoint (required for ollama, optional for openai)
+	AutonomousMode bool   `json:"autonomous_mode"` // when true, AI executes commands without approval
+	CustomContext  string `json:"custom_context"`  // user-provided context about their infrastructure
+}
+
+// AIProvider constants
+const (
+	AIProviderAnthropic = "anthropic"
+	AIProviderOpenAI    = "openai"
+	AIProviderOllama    = "ollama"
+)
+
+// Default models per provider
+const (
+	DefaultAIModelAnthropic = "claude-opus-4-5-20251101"
+	DefaultAIModelOpenAI    = "gpt-4o"
+	DefaultAIModelOllama    = "llama3"
+	DefaultOllamaBaseURL    = "http://localhost:11434"
+)
+
+// NewDefaultAIConfig returns an AIConfig with sensible defaults
+func NewDefaultAIConfig() *AIConfig {
+	return &AIConfig{
+		Enabled:  false,
+		Provider: AIProviderAnthropic,
+		Model:    DefaultAIModelAnthropic,
+	}
+}
+
+// IsConfigured returns true if the AI config has enough info to make API calls
+func (c *AIConfig) IsConfigured() bool {
+	if !c.Enabled {
+		return false
+	}
+
+	switch c.Provider {
+	case AIProviderAnthropic, AIProviderOpenAI:
+		return c.APIKey != ""
+	case AIProviderOllama:
+		// Ollama doesn't need an API key
+		return true
+	default:
+		return false
+	}
+}
+
+// GetBaseURL returns the base URL, using defaults where appropriate
+func (c *AIConfig) GetBaseURL() string {
+	if c.BaseURL != "" {
+		return c.BaseURL
+	}
+	if c.Provider == AIProviderOllama {
+		return DefaultOllamaBaseURL
+	}
+	return ""
+}
+
+// GetModel returns the model, using defaults where appropriate
+func (c *AIConfig) GetModel() string {
+	if c.Model != "" {
+		return c.Model
+	}
+	switch c.Provider {
+	case AIProviderAnthropic:
+		return DefaultAIModelAnthropic
+	case AIProviderOpenAI:
+		return DefaultAIModelOpenAI
+	case AIProviderOllama:
+		return DefaultAIModelOllama
+	default:
+		return ""
+	}
+}
--- a/internal/config/docker_metadata.go
+++ b/internal/config/docker_metadata.go
@@ -16,6 +16,7 @@ type DockerMetadata struct {
 	CustomURL   string   `json:"customUrl"`   // Custom URL for the resource
 	Description string   `json:"description"` // Optional description
 	Tags        []string `json:"tags"`        // Optional tags for categorization
+	Notes       []string `json:"notes"`       // User annotations for AI context
 }

 // DockerHostMetadata holds additional metadata for a Docker host
--- a/internal/config/guest_metadata.go
+++ b/internal/config/guest_metadata.go
@@ -16,6 +16,7 @@ type GuestMetadata struct {
 	CustomURL   string   `json:"customUrl"`   // Custom URL for the guest
 	Description string   `json:"description"` // Optional description
 	Tags        []string `json:"tags"`        // Optional tags for categorization
+	Notes       []string `json:"notes"`       // User annotations for AI context (e.g., "Runs PBS in Docker")
 	// Last-known identity (persisted even after guest deletion)
 	LastKnownName string `json:"lastKnownName,omitempty"` // Last known guest name
 	LastKnownType string `json:"lastKnownType,omitempty"` // Last known guest type (qemu, lxc)
--- a/internal/config/persistence.go
+++ b/internal/config/persistence.go
@@ -31,6 +31,7 @@ type ConfigPersistence struct {
 	systemFile    string
 	oidcFile      string
 	apiTokensFile string
+	aiFile        string
 	crypto        *crypto.CryptoManager
 }

@@ -73,6 +74,7 @@ func newConfigPersistence(configDir string) (*ConfigPersistence, error) {
 		systemFile:    filepath.Join(configDir, "system.json"),
 		oidcFile:      filepath.Join(configDir, "oidc.enc"),
 		apiTokensFile: filepath.Join(configDir, "api_tokens.json"),
+		aiFile:        filepath.Join(configDir, "ai.enc"),
 		crypto:        cryptoMgr,
 	}

@@ -1291,6 +1293,67 @@ func (c *ConfigPersistence) LoadOIDCConfig() (*OIDCConfig, error) {
 	return &settings, nil
 }

+// SaveAIConfig stores AI settings, encrypting them when a crypto manager is available.
+func (c *ConfigPersistence) SaveAIConfig(settings AIConfig) error {
+	c.mu.Lock()
+	defer c.mu.Unlock()
+
+	if err := c.EnsureConfigDir(); err != nil {
+		return err
+	}
+
+	data, err := json.Marshal(settings)
+	if err != nil {
+		return err
+	}
+
+	if c.crypto != nil {
+		encrypted, err := c.crypto.Encrypt(data)
+		if err != nil {
+			return err
+		}
+		data = encrypted
+	}
+
+	if err := c.writeConfigFileLocked(c.aiFile, data, 0600); err != nil {
+		return err
+	}
+
+	log.Info().Str("file", c.aiFile).Bool("enabled", settings.Enabled).Msg("AI configuration saved")
+	return nil
+}
+
+// LoadAIConfig retrieves the persisted AI settings. It returns default config when no configuration exists yet.
+func (c *ConfigPersistence) LoadAIConfig() (*AIConfig, error) {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	data, err := os.ReadFile(c.aiFile)
+	if err != nil {
+		if os.IsNotExist(err) {
+			// Return default config if file doesn't exist
+			return NewDefaultAIConfig(), nil
+		}
+		return nil, err
+	}
+
+	if c.crypto != nil {
+		decrypted, err := c.crypto.Decrypt(data)
+		if err != nil {
+			return nil, err
+		}
+		data = decrypted
+	}
+
+	var settings AIConfig
+	if err := json.Unmarshal(data, &settings); err != nil {
+		return nil, err
+	}
+
+	log.Info().Str("file", c.aiFile).Bool("enabled", settings.Enabled).Msg("AI configuration loaded")
+	return &settings, nil
+}
+
 // LoadSystemSettings loads system settings from file
 func (c *ConfigPersistence) LoadSystemSettings() (*SystemSettings, error) {
 	c.mu.RLock()
@@ -1440,3 +1503,60 @@ func (c *ConfigPersistence) cleanupOldBackups(pattern string) {
 		}
 	}
 }
+
+// LoadGuestMetadata loads all guest metadata from disk (for AI context)
+func (c *ConfigPersistence) LoadGuestMetadata() (map[string]*GuestMetadata, error) {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	filePath := filepath.Join(c.configDir, "guest_metadata.json")
+	data, err := os.ReadFile(filePath)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return make(map[string]*GuestMetadata), nil
+		}
+		return nil, err
+	}
+
+	var metadata map[string]*GuestMetadata
+	if err := json.Unmarshal(data, &metadata); err != nil {
+		return nil, err
+	}
+
+	return metadata, nil
+}
+
+// LoadDockerMetadata loads all docker metadata from disk (for AI context)
+func (c *ConfigPersistence) LoadDockerMetadata() (map[string]*DockerMetadata, error) {
+	c.mu.RLock()
+	defer c.mu.RUnlock()
+
+	filePath := filepath.Join(c.configDir, "docker_metadata.json")
+	data, err := os.ReadFile(filePath)
+	if err != nil {
+		if os.IsNotExist(err) {
+			return make(map[string]*DockerMetadata), nil
+		}
+		return nil, err
+	}
+
+	// Try versioned format first
+	var fileData struct {
+		Containers map[string]*DockerMetadata `json:"containers,omitempty"`
+	}
+	if err := json.Unmarshal(data, &fileData); err != nil {
+		return nil, err
+	}
+
+	if fileData.Containers != nil {
+		return fileData.Containers, nil
+	}
+
+	// Fall back to legacy format (direct map)
+	var metadata map[string]*DockerMetadata
+	if err := json.Unmarshal(data, &metadata); err != nil {
+		return nil, err
+	}
+
+	return metadata, nil
+}
--- a/internal/hostagent/agent.go
+++ b/internal/hostagent/agent.go
@@ -57,6 +57,7 @@ type Agent struct {
 	interval        time.Duration
 	trimmedPulseURL string
 	reportBuffer    *buffer.Queue[agentshost.Report]
+	commandClient   *CommandClient
 }

 const defaultInterval = 30 * time.Second
@@ -168,7 +169,7 @@ func New(cfg Config) (*Agent, error) {

 	const bufferCapacity = 60

-	return &Agent{
+	agent := &Agent{
 		cfg:             cfg,
 		logger:          logger,
 		httpClient:      client,
@@ -186,7 +187,12 @@ func New(cfg Config) (*Agent, error) {
 		interval:        cfg.Interval,
 		trimmedPulseURL: pulseURL,
 		reportBuffer:    buffer.New[agentshost.Report](bufferCapacity),
-	}, nil
+	}
+
+	// Create command client for AI command execution
+	agent.commandClient = NewCommandClient(cfg, agentID, hostname, platform, agentVersion)
+
+	return agent, nil
 }

 // Run executes the agent until the context is cancelled.
@@ -195,6 +201,15 @@ func (a *Agent) Run(ctx context.Context) error {
 		return a.runOnce(ctx)
 	}

+	// Start command client in background for AI command execution
+	if a.commandClient != nil {
+		go func() {
+			if err := a.commandClient.Run(ctx); err != nil && !errors.Is(err, context.Canceled) {
+				a.logger.Error().Err(err).Msg("Command client stopped with error")
+			}
+		}()
+	}
+
 	ticker := time.NewTicker(a.interval)
 	defer ticker.Stop()

--- a/internal/hostagent/commands.go
+++ b/internal/hostagent/commands.go
@@ -0,0 +1,437 @@
+package hostagent
+
+import (
+	"bytes"
+	"context"
+	"crypto/tls"
+	"encoding/json"
+	"fmt"
+	"net/url"
+	"os/exec"
+	"runtime"
+	"strings"
+	"sync"
+	"time"
+
+	"github.com/gorilla/websocket"
+	"github.com/rs/zerolog"
+)
+
+// CommandClient handles WebSocket connection to Pulse for AI command execution
+type CommandClient struct {
+	pulseURL           string
+	apiToken           string
+	agentID            string
+	hostname           string
+	platform           string
+	version            string
+	insecureSkipVerify bool
+	logger             zerolog.Logger
+
+	conn     *websocket.Conn
+	connMu   sync.Mutex
+	done     chan struct{}
+}
+
+// NewCommandClient creates a new command execution client
+func NewCommandClient(cfg Config, agentID, hostname, platform, version string) *CommandClient {
+	logger := cfg.Logger.With().Str("component", "command-client").Logger()
+
+	return &CommandClient{
+		pulseURL:           strings.TrimRight(cfg.PulseURL, "/"),
+		apiToken:           cfg.APIToken,
+		agentID:            agentID,
+		hostname:           hostname,
+		platform:           platform,
+		version:            version,
+		insecureSkipVerify: cfg.InsecureSkipVerify,
+		logger:             logger,
+		done:               make(chan struct{}),
+	}
+}
+
+// Message types matching agentexec package
+type messageType string
+
+const (
+	msgTypeAgentRegister messageType = "agent_register"
+	msgTypeAgentPing     messageType = "agent_ping"
+	msgTypeCommandResult messageType = "command_result"
+	msgTypeRegistered    messageType = "registered"
+	msgTypePong          messageType = "pong"
+	msgTypeExecuteCmd    messageType = "execute_command"
+	msgTypeReadFile      messageType = "read_file"
+)
+
+type wsMessage struct {
+	Type      messageType     `json:"type"`
+	ID        string          `json:"id,omitempty"`
+	Timestamp time.Time       `json:"timestamp"`
+	Payload   json.RawMessage `json:"payload,omitempty"`
+}
+
+type registerPayload struct {
+	AgentID  string   `json:"agent_id"`
+	Hostname string   `json:"hostname"`
+	Version  string   `json:"version"`
+	Platform string   `json:"platform"`
+	Tags     []string `json:"tags,omitempty"`
+	Token    string   `json:"token"`
+}
+
+type registeredPayload struct {
+	Success bool   `json:"success"`
+	Message string `json:"message,omitempty"`
+}
+
+type executeCommandPayload struct {
+	RequestID  string `json:"request_id"`
+	Command    string `json:"command"`
+	TargetType string `json:"target_type"`
+	TargetID   string `json:"target_id,omitempty"`
+	Timeout    int    `json:"timeout,omitempty"`
+}
+
+type commandResultPayload struct {
+	RequestID string `json:"request_id"`
+	Success   bool   `json:"success"`
+	Stdout    string `json:"stdout,omitempty"`
+	Stderr    string `json:"stderr,omitempty"`
+	ExitCode  int    `json:"exit_code"`
+	Error     string `json:"error,omitempty"`
+	Duration  int64  `json:"duration_ms"`
+}
+
+// Run starts the command client and maintains the WebSocket connection
+func (c *CommandClient) Run(ctx context.Context) error {
+	for {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		default:
+		}
+
+		if err := c.connectAndHandle(ctx); err != nil {
+			if ctx.Err() != nil {
+				return ctx.Err()
+			}
+			c.logger.Warn().Err(err).Msg("WebSocket connection failed, reconnecting in 10s")
+			select {
+			case <-ctx.Done():
+				return ctx.Err()
+			case <-time.After(10 * time.Second):
+			}
+		}
+	}
+}
+
+func (c *CommandClient) connectAndHandle(ctx context.Context) error {
+	// Build WebSocket URL
+	wsURL, err := c.buildWebSocketURL()
+	if err != nil {
+		return fmt.Errorf("build websocket url: %w", err)
+	}
+
+	c.logger.Info().Str("url", wsURL).Msg("Connecting to Pulse command server")
+
+	// Create dialer with TLS config
+	dialer := websocket.Dialer{
+		TLSClientConfig: &tls.Config{
+			MinVersion:         tls.VersionTLS12,
+			InsecureSkipVerify: c.insecureSkipVerify,
+		},
+		HandshakeTimeout: 10 * time.Second,
+	}
+
+	// Connect
+	conn, _, err := dialer.DialContext(ctx, wsURL, nil)
+	if err != nil {
+		return fmt.Errorf("dial websocket: %w", err)
+	}
+
+	c.connMu.Lock()
+	c.conn = conn
+	c.connMu.Unlock()
+
+	defer func() {
+		c.connMu.Lock()
+		c.conn = nil
+		c.connMu.Unlock()
+		conn.Close()
+	}()
+
+	// Send registration
+	if err := c.sendRegistration(conn); err != nil {
+		return fmt.Errorf("send registration: %w", err)
+	}
+
+	// Wait for registration response
+	if err := c.waitForRegistration(conn); err != nil {
+		return fmt.Errorf("registration failed: %w", err)
+	}
+
+	c.logger.Info().Msg("Connected and registered with Pulse command server")
+
+	// Start ping loop
+	pingDone := make(chan struct{})
+	go c.pingLoop(ctx, conn, pingDone)
+	defer close(pingDone)
+
+	// Handle incoming messages
+	return c.handleMessages(ctx, conn)
+}
+
+func (c *CommandClient) buildWebSocketURL() (string, error) {
+	parsed, err := url.Parse(c.pulseURL)
+	if err != nil {
+		return "", err
+	}
+
+	// Convert http(s) to ws(s)
+	switch parsed.Scheme {
+	case "https":
+		parsed.Scheme = "wss"
+	case "http":
+		parsed.Scheme = "ws"
+	case "wss", "ws":
+		// Already WebSocket scheme
+	default:
+		parsed.Scheme = "ws"
+	}
+
+	parsed.Path = "/api/agent/ws"
+	return parsed.String(), nil
+}
+
+func (c *CommandClient) sendRegistration(conn *websocket.Conn) error {
+	payload, _ := json.Marshal(registerPayload{
+		AgentID:  c.agentID,
+		Hostname: c.hostname,
+		Version:  c.version,
+		Platform: c.platform,
+		Token:    c.apiToken,
+	})
+
+	msg := wsMessage{
+		Type:      msgTypeAgentRegister,
+		Timestamp: time.Now(),
+		Payload:   payload,
+	}
+
+	return conn.WriteJSON(msg)
+}
+
+func (c *CommandClient) waitForRegistration(conn *websocket.Conn) error {
+	conn.SetReadDeadline(time.Now().Add(30 * time.Second))
+	defer conn.SetReadDeadline(time.Time{})
+
+	var msg wsMessage
+	if err := conn.ReadJSON(&msg); err != nil {
+		return fmt.Errorf("read registration response: %w", err)
+	}
+
+	if msg.Type != msgTypeRegistered {
+		return fmt.Errorf("unexpected message type: %s", msg.Type)
+	}
+
+	var payload registeredPayload
+	if err := json.Unmarshal(msg.Payload, &payload); err != nil {
+		return fmt.Errorf("parse registration response: %w", err)
+	}
+
+	if !payload.Success {
+		return fmt.Errorf("registration rejected: %s", payload.Message)
+	}
+
+	return nil
+}
+
+func (c *CommandClient) pingLoop(ctx context.Context, conn *websocket.Conn, done chan struct{}) {
+	ticker := time.NewTicker(30 * time.Second)
+	defer ticker.Stop()
+
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-done:
+			return
+		case <-ticker.C:
+			c.connMu.Lock()
+			if c.conn != nil {
+				msg := wsMessage{
+					Type:      msgTypeAgentPing,
+					Timestamp: time.Now(),
+				}
+				if err := conn.WriteJSON(msg); err != nil {
+					c.logger.Debug().Err(err).Msg("Failed to send ping")
+				}
+			}
+			c.connMu.Unlock()
+		}
+	}
+}
+
+func (c *CommandClient) handleMessages(ctx context.Context, conn *websocket.Conn) error {
+	for {
+		select {
+		case <-ctx.Done():
+			return ctx.Err()
+		default:
+		}
+
+		var msg wsMessage
+		if err := conn.ReadJSON(&msg); err != nil {
+			return fmt.Errorf("read message: %w", err)
+		}
+
+		switch msg.Type {
+		case msgTypePong:
+			// Ignore pong responses
+
+		case msgTypeExecuteCmd:
+			var payload executeCommandPayload
+			if err := json.Unmarshal(msg.Payload, &payload); err != nil {
+				c.logger.Error().Err(err).Msg("Failed to parse execute_command payload")
+				continue
+			}
+
+			// Execute command in background
+			go c.handleExecuteCommand(ctx, conn, payload)
+
+		case msgTypeReadFile:
+			// Handle read_file similarly (uses cat command internally)
+			var payload executeCommandPayload
+			if err := json.Unmarshal(msg.Payload, &payload); err != nil {
+				c.logger.Error().Err(err).Msg("Failed to parse read_file payload")
+				continue
+			}
+			go c.handleExecuteCommand(ctx, conn, payload)
+
+		default:
+			c.logger.Debug().Str("type", string(msg.Type)).Msg("Unknown message type")
+		}
+	}
+}
+
+func (c *CommandClient) handleExecuteCommand(ctx context.Context, conn *websocket.Conn, payload executeCommandPayload) {
+	startTime := time.Now()
+
+	c.logger.Info().
+		Str("request_id", payload.RequestID).
+		Str("command", payload.Command).
+		Str("target_type", payload.TargetType).
+		Str("target_id", payload.TargetID).
+		Msg("Executing command")
+
+	result := c.executeCommand(ctx, payload)
+	result.Duration = time.Since(startTime).Milliseconds()
+
+	// Send result back
+	resultPayload, _ := json.Marshal(result)
+	msg := wsMessage{
+		Type:      msgTypeCommandResult,
+		ID:        payload.RequestID,
+		Timestamp: time.Now(),
+		Payload:   resultPayload,
+	}
+
+	c.connMu.Lock()
+	err := conn.WriteJSON(msg)
+	c.connMu.Unlock()
+
+	if err != nil {
+		c.logger.Error().Err(err).Str("request_id", payload.RequestID).Msg("Failed to send command result")
+	} else {
+		c.logger.Info().
+			Str("request_id", payload.RequestID).
+			Bool("success", result.Success).
+			Int("exit_code", result.ExitCode).
+			Int64("duration_ms", result.Duration).
+			Msg("Command completed")
+	}
+}
+
+func (c *CommandClient) executeCommand(ctx context.Context, payload executeCommandPayload) commandResultPayload {
+	result := commandResultPayload{
+		RequestID: payload.RequestID,
+	}
+
+	// Determine timeout
+	timeout := time.Duration(payload.Timeout) * time.Second
+	if timeout <= 0 {
+		timeout = 60 * time.Second
+	}
+
+	cmdCtx, cancel := context.WithTimeout(ctx, timeout)
+	defer cancel()
+
+	// Build the command based on target type
+	command := payload.Command
+
+	// If targeting a container or VM, wrap the command
+	if payload.TargetType == "container" && payload.TargetID != "" {
+		// Use pct exec for LXC containers
+		command = fmt.Sprintf("pct exec %s -- %s", payload.TargetID, payload.Command)
+	} else if payload.TargetType == "vm" && payload.TargetID != "" {
+		// Use qm guest exec for VMs (requires QEMU guest agent)
+		command = fmt.Sprintf("qm guest exec %s -- %s", payload.TargetID, payload.Command)
+	}
+
+	// Execute the command
+	var cmd *exec.Cmd
+	if runtime.GOOS == "windows" {
+		cmd = exec.CommandContext(cmdCtx, "cmd", "/C", command)
+	} else {
+		cmd = exec.CommandContext(cmdCtx, "sh", "-c", command)
+	}
+
+	var stdout, stderr bytes.Buffer
+	cmd.Stdout = &stdout
+	cmd.Stderr = &stderr
+
+	err := cmd.Run()
+
+	result.Stdout = stdout.String()
+	result.Stderr = stderr.String()
+
+	if err != nil {
+		if exitErr, ok := err.(*exec.ExitError); ok {
+			result.ExitCode = exitErr.ExitCode()
+			result.Success = false
+		} else if cmdCtx.Err() == context.DeadlineExceeded {
+			result.Error = "command timed out"
+			result.ExitCode = -1
+			result.Success = false
+		} else {
+			result.Error = err.Error()
+			result.ExitCode = -1
+			result.Success = false
+		}
+	} else {
+		result.ExitCode = 0
+		result.Success = true
+	}
+
+	// Truncate output if too large (1MB limit)
+	const maxOutputSize = 1024 * 1024
+	if len(result.Stdout) > maxOutputSize {
+		result.Stdout = result.Stdout[:maxOutputSize] + "\n... (output truncated)"
+	}
+	if len(result.Stderr) > maxOutputSize {
+		result.Stderr = result.Stderr[:maxOutputSize] + "\n... (output truncated)"
+	}
+
+	return result
+}
+
+// Close closes the command client connection
+func (c *CommandClient) Close() error {
+	c.connMu.Lock()
+	defer c.connMu.Unlock()
+
+	if c.conn != nil {
+		return c.conn.Close()
+	}
+	return nil
+}
--- a/mock.env
+++ b/mock.env
@@ -1,4 +1,4 @@
-PULSE_MOCK_MODE=true
+PULSE_MOCK_MODE=false
 PULSE_MOCK_NODES=7
 PULSE_MOCK_VMS_PER_NODE=5
 PULSE_MOCK_LXCS_PER_NODE=8
--- a/scripts/hot-dev.sh
+++ b/scripts/hot-dev.sh
@@ -78,7 +78,9 @@ if [[ -z ${PULSE_DEV_WS_URL:-} ]]; then
    fi
 fi

-ALLOWED_ORIGINS="*"
+# Set specific allowed origin for CORS with credentials
+# Use the frontend dev URL so cross-port SSE requests work with auth cookies
+ALLOWED_ORIGINS="http://${PULSE_DEV_API_HOST:-127.0.0.1}:${FRONTEND_DEV_PORT:-7655}"

 export FRONTEND_PORT PORT
 export FRONTEND_DEV_HOST FRONTEND_DEV_PORT
--- a/scripts/install.sh
+++ b/scripts/install.sh
@@ -325,18 +325,26 @@ chmod +x "${INSTALL_DIR}/${BINARY_NAME}"

 # --- Legacy Cleanup ---
 # Remove old agents if they exist to prevent conflicts
+# This is critical because legacy agents use the same system ID and will cause
+# connection conflicts (rapid connect/disconnect cycles) with the unified agent.
 log_info "Checking for legacy agents..."

-# Legacy Host Agent
+# Kill any running legacy agent processes first (even if started manually)
+# This prevents WebSocket connection conflicts during installation
+pkill -f "pulse-host-agent" 2>/dev/null || true
+pkill -f "pulse-docker-agent" 2>/dev/null || true
+sleep 1
+
+# Legacy Host Agent - systemd cleanup
 if command -v systemctl >/dev/null 2>&1; then
-    if systemctl is-active --quiet pulse-host-agent 2>/dev/null || systemctl is-enabled --quiet pulse-host-agent 2>/dev/null; then
+    if systemctl is-active --quiet pulse-host-agent 2>/dev/null || systemctl is-enabled --quiet pulse-host-agent 2>/dev/null || [[ -f /etc/systemd/system/pulse-host-agent.service ]]; then
        log_warn "Removing legacy pulse-host-agent..."
        systemctl stop pulse-host-agent 2>/dev/null || true
        systemctl disable pulse-host-agent 2>/dev/null || true
        rm -f /etc/systemd/system/pulse-host-agent.service
        rm -f /usr/local/bin/pulse-host-agent
    fi
-    if systemctl is-active --quiet pulse-docker-agent 2>/dev/null || systemctl is-enabled --quiet pulse-docker-agent 2>/dev/null; then
+    if systemctl is-active --quiet pulse-docker-agent 2>/dev/null || systemctl is-enabled --quiet pulse-docker-agent 2>/dev/null || [[ -f /etc/systemd/system/pulse-docker-agent.service ]]; then
        log_warn "Removing legacy pulse-docker-agent..."
        systemctl stop pulse-docker-agent 2>/dev/null || true
        systemctl disable pulse-docker-agent 2>/dev/null || true