diff --git a/backend/internal/domain/settings/handler.go b/backend/internal/domain/settings/handler.go
new file mode 100644
index 0000000..e3aefe3
--- /dev/null
+++ b/backend/internal/domain/settings/handler.go
@@ -0,0 +1,75 @@
+package settings
+
+import (
+	"net/http"
+
+	"github.com/gin-gonic/gin"
+
+	"marktvogt.de/backend/internal/pkg/ai"
+)
+
+// AIStatus is the response payload for GET /admin/settings/ai.
+type AIStatus struct {
+	Provider  string   `json:"provider"`
+	Connected bool     `json:"connected"`
+	BaseURL   string   `json:"base_url,omitempty"`
+	Model     string   `json:"model"`
+	Models    []string `json:"models"`
+}
+
+// Handler serves AI settings endpoints. ollama is nil when the active
+// provider is not Ollama.
+type Handler struct {
+	ollama   *ai.OllamaProvider
+	provider string
+}
+
+func NewHandler(provider ai.Provider) *Handler {
+	ollama, _ := provider.(*ai.OllamaProvider)
+	return &Handler{ollama: ollama, provider: provider.Name()}
+}
+
+func (h *Handler) GetAI(c *gin.Context) {
+	if h.ollama == nil {
+		c.JSON(http.StatusOK, gin.H{"data": AIStatus{
+			Provider:  h.provider,
+			Connected: true,
+			Model:     "",
+			Models:    []string{},
+		}})
+		return
+	}
+
+	models, err := h.ollama.ListModels(c.Request.Context())
+	status := AIStatus{
+		Provider: "ollama",
+		BaseURL:  h.ollama.BaseURL(),
+		Model:    h.ollama.Model(),
+		Models:   []string{},
+	}
+	if err != nil {
+		status.Connected = false
+	} else {
+		status.Connected = true
+		for _, m := range models {
+			status.Models = append(status.Models, m.Name)
+		}
+	}
+	c.JSON(http.StatusOK, gin.H{"data": status})
+}
+
+func (h *Handler) SetModel(c *gin.Context) {
+	if h.ollama == nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "model switching only supported for Ollama provider"})
+		return
+	}
+	var req struct {
+		Model string `json:"model" binding:"required"`
+	}
+	if err := c.ShouldBindJSON(&req); err != nil {
+		c.JSON(http.StatusBadRequest, gin.H{"error": "model is required"})
+		return
+	}
+	h.ollama.SetModel(req.Model)
+	c.JSON(http.StatusOK, gin.H{"data": gin.H{"model": req.Model}})
+}
diff --git a/backend/internal/domain/settings/routes.go b/backend/internal/domain/settings/routes.go
new file mode 100644
index 0000000..160c56e
--- /dev/null
+++ b/backend/internal/domain/settings/routes.go
@@ -0,0 +1,9 @@
+package settings
+
+import "github.com/gin-gonic/gin"
+
+func RegisterRoutes(rg *gin.RouterGroup, h *Handler, requireAuth, requireAdmin gin.HandlerFunc) {
+	admin := rg.Group("/admin", requireAuth, requireAdmin)
+	admin.GET("/settings/ai", h.GetAI)
+	admin.POST("/settings/ai/model", h.SetModel)
+}
diff --git a/backend/internal/pkg/ai/ollama.go b/backend/internal/pkg/ai/ollama.go
index 95ee20d..dcc7781 100644
--- a/backend/internal/pkg/ai/ollama.go
+++ b/backend/internal/pkg/ai/ollama.go
@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"sync"
 	"time"
 )
 
@@ -17,8 +18,10 @@ type OllamaConfig struct {
 }
 
 type OllamaProvider struct {
-	cfg    OllamaConfig
-	client *http.Client
+	cfg         OllamaConfig
+	client      *http.Client
+	mu          sync.RWMutex
+	activeModel string
 }
 
 func NewOllamaProvider(cfg OllamaConfig) *OllamaProvider {
@@ -26,8 +29,9 @@ func NewOllamaProvider(cfg OllamaConfig) *OllamaProvider {
 		cfg.Timeout = 300 * time.Second
 	}
 	return &OllamaProvider{
-		cfg:    cfg,
-		client: &http.Client{Timeout: cfg.Timeout},
+		cfg:         cfg,
+		client:      &http.Client{Timeout: cfg.Timeout},
+		activeModel: cfg.Model,
 	}
 }
 
@@ -35,6 +39,59 @@ func (p *OllamaProvider) Name() string             { return "ollama" }
 func (p *OllamaProvider) SupportsJSONMode() bool   { return true }
 func (p *OllamaProvider) SupportsJSONSchema() bool { return true }
 
+func (p *OllamaProvider) BaseURL() string {
+	return p.cfg.BaseURL
+}
+
+func (p *OllamaProvider) Model() string {
+	p.mu.RLock()
+	defer p.mu.RUnlock()
+	return p.activeModel
+}
+
+func (p *OllamaProvider) SetModel(model string) {
+	p.mu.Lock()
+	defer p.mu.Unlock()
+	p.activeModel = model
+}
+
+// OllamaModelInfo is a model entry from Ollama's /api/tags response.
+type OllamaModelInfo struct {
+	Name string `json:"name"`
+	Size int64  `json:"size"`
+}
+
+// ListModels calls Ollama's /api/tags endpoint and returns available models.
+func (p *OllamaProvider) ListModels(ctx context.Context) ([]OllamaModelInfo, error) {
+	req, err := http.NewRequestWithContext(ctx, http.MethodGet, p.cfg.BaseURL+"/api/tags", nil)
+	if err != nil {
+		return nil, err
+	}
+	resp, err := p.client.Do(req)
+	if err != nil {
+		return nil, err
+	}
+	defer func() { _ = resp.Body.Close() }()
+	if resp.StatusCode != http.StatusOK {
+		b, _ := io.ReadAll(resp.Body)
+		return nil, fmt.Errorf("ollama /api/tags: status %d: %s", resp.StatusCode, b)
+	}
+	var body struct {
+		Models []struct {
+			Name string `json:"name"`
+			Size int64  `json:"size"`
+		} `json:"models"`
+	}
+	if err := json.NewDecoder(resp.Body).Decode(&body); err != nil {
+		return nil, err
+	}
+	out := make([]OllamaModelInfo, len(body.Models))
+	for i, m := range body.Models {
+		out[i] = OllamaModelInfo{Name: m.Name, Size: m.Size}
+	}
+	return out, nil
+}
+
 type ollamaChatReq struct {
 	Model    string          `json:"model"`
 	Messages []ollamaMessage `json:"messages"`
@@ -64,7 +121,7 @@ type ollamaChatResp struct {
 func (p *OllamaProvider) Chat(ctx context.Context, req *ChatRequest) (*ChatResponse, error) {
 	model := req.Model
 	if model == "" {
-		model = p.cfg.Model
+		model = p.Model()
 	}
 	body := ollamaChatReq{
 		Model:    model,
diff --git a/backend/internal/server/routes.go b/backend/internal/server/routes.go
index a8e34f0..b4a49e5 100644
--- a/backend/internal/server/routes.go
+++ b/backend/internal/server/routes.go
@@ -11,6 +11,7 @@ import (
 	"marktvogt.de/backend/internal/domain/discovery/crawler"
 	"marktvogt.de/backend/internal/domain/discovery/enrich"
 	"marktvogt.de/backend/internal/domain/market"
+	"marktvogt.de/backend/internal/domain/settings"
 	"marktvogt.de/backend/internal/domain/user"
 	"marktvogt.de/backend/internal/middleware"
 	"marktvogt.de/backend/internal/pkg/ai"
@@ -89,6 +90,10 @@ func (s *Server) registerRoutes() {
 	discoveryHandler := discovery.NewHandler(discoveryService, s.cfg.Discovery.CrawlerManualRateLimitPerHour)
 	requireTickToken := middleware.RequireBearerToken(s.cfg.Discovery.Token)
 	discovery.RegisterRoutes(v1, discoveryHandler, requireAuth, requireAdmin, requireTickToken)
+
+	// AI settings routes
+	settingsHandler := settings.NewHandler(aiProvider)
+	settings.RegisterRoutes(v1, settingsHandler, requireAuth, requireAdmin)
 }
 
 func (s *Server) healthz(c *gin.Context) {
diff --git a/web/src/lib/api/types.ts b/web/src/lib/api/types.ts
index 98dd3cd..460edf0 100644
--- a/web/src/lib/api/types.ts
+++ b/web/src/lib/api/types.ts
@@ -183,6 +183,15 @@ export interface SubmitMarketRequest {
 	turnstile_token: string;
 }
 
+// AI settings
+export interface AIStatus {
+	provider: string;
+	connected: boolean;
+	base_url?: string;
+	model: string;
+	models: string[];
+}
+
 // AI Research types
 export interface ResearchResult {
 	suggestions: FieldSuggestion[];
diff --git a/web/src/routes/admin/+layout.svelte b/web/src/routes/admin/+layout.svelte
index b7c6998..b07a3e4 100644
--- a/web/src/routes/admin/+layout.svelte
+++ b/web/src/routes/admin/+layout.svelte
@@ -10,7 +10,8 @@
 
 	const navItems = [
 		{ href: '/admin/maerkte', label: 'Märkte' },
-		{ href: '/admin/discovery', label: 'Discovery' }
+		{ href: '/admin/discovery', label: 'Discovery' },
+		{ href: '/admin/einstellungen', label: 'Einstellungen' }
 	];
 
 	function isActive(href: string): boolean {
diff --git a/web/src/routes/admin/einstellungen/+page.server.ts b/web/src/routes/admin/einstellungen/+page.server.ts
new file mode 100644
index 0000000..01b4f77
--- /dev/null
+++ b/web/src/routes/admin/einstellungen/+page.server.ts
@@ -0,0 +1,30 @@
+import { fail } from '@sveltejs/kit';
+import { serverFetch } from '$lib/api/client.server.js';
+import type { AIStatus } from '$lib/api/types.js';
+import type { Actions, PageServerLoad } from './$types.js';
+
+export const load: PageServerLoad = async ({ cookies, fetch }) => {
+	const res = await serverFetch<AIStatus>('/admin/settings/ai', cookies, { fetch });
+	return { ai: res.data };
+};
+
+export const actions: Actions = {
+	setModel: async ({ cookies, fetch, request }) => {
+		const data = await request.formData();
+		const model = data.get('model');
+		if (!model || typeof model !== 'string') {
+			return fail(400, { error: 'Kein Modell ausgewählt.' });
+		}
+		try {
+			await serverFetch('/admin/settings/ai/model', cookies, {
+				method: 'POST',
+				body: JSON.stringify({ model }),
+				fetch
+			});
+			return { success: true, model };
+		} catch (err) {
+			const message = err instanceof Error ? err.message : 'Modell konnte nicht gesetzt werden.';
+			return fail(500, { error: message });
+		}
+	}
+};
diff --git a/web/src/routes/admin/einstellungen/+page.svelte b/web/src/routes/admin/einstellungen/+page.svelte
new file mode 100644
index 0000000..46e8aad
--- /dev/null
+++ b/web/src/routes/admin/einstellungen/+page.svelte
@@ -0,0 +1,136 @@
+<script lang="ts">
+	import { untrack } from 'svelte';
+	import { enhance } from '$app/forms';
+	import type { PageData, ActionData } from './$types.js';
+
+	interface Props {
+		data: PageData;
+		form: ActionData;
+	}
+
+	let { data, form }: Props = $props();
+
+	let selectedModel = $state(untrack(() => data.ai.model));
+	let saving = $state(false);
+
+	let activeModel = $derived(form?.success && form.model ? form.model : data.ai.model);
+</script>
+
+<div class="space-y-6">
+	<div>
+		<h1 class="text-2xl font-bold text-stone-900 dark:text-stone-100">Einstellungen</h1>
+		<p class="mt-1 text-sm text-stone-500 dark:text-stone-400">
+			Systemkonfiguration und KI-Provider
+		</p>
+	</div>
+
+	<div class="rounded-lg border border-stone-200 bg-white dark:border-stone-700 dark:bg-stone-900">
+		<div class="border-b border-stone-200 px-6 py-4 dark:border-stone-700">
+			<h2 class="text-base font-semibold text-stone-900 dark:text-stone-100">KI-Provider</h2>
+		</div>
+
+		<div class="space-y-4 px-6 py-4">
+			<!-- Provider + status row -->
+			<div class="flex items-center gap-4">
+				<div>
+					<span class="text-xs font-medium tracking-wide text-stone-400 uppercase">Provider</span>
+					<p class="mt-0.5 font-mono text-sm font-semibold text-stone-800 dark:text-stone-200">
+						{data.ai.provider}
+					</p>
+				</div>
+
+				{#if data.ai.base_url}
+					<div>
+						<span class="text-xs font-medium tracking-wide text-stone-400 uppercase">URL</span>
+						<p class="mt-0.5 font-mono text-sm text-stone-600 dark:text-stone-400">
+							{data.ai.base_url}
+						</p>
+					</div>
+				{/if}
+
+				<div class="ml-auto">
+					{#if data.ai.connected}
+						<span
+							class="inline-flex items-center gap-1.5 rounded-full bg-green-100 px-3 py-1 text-xs font-medium text-green-800 dark:bg-green-900 dark:text-green-200"
+						>
+							<span class="h-1.5 w-1.5 rounded-full bg-green-500"></span>
+							Verbunden
+						</span>
+					{:else}
+						<span
+							class="inline-flex items-center gap-1.5 rounded-full bg-red-100 px-3 py-1 text-xs font-medium text-red-800 dark:bg-red-900 dark:text-red-200"
+						>
+							<span class="h-1.5 w-1.5 rounded-full bg-red-500"></span>
+							Nicht erreichbar
+						</span>
+					{/if}
+				</div>
+			</div>
+
+			<!-- Model selector (Ollama only) -->
+			{#if data.ai.provider === 'ollama'}
+				<div class="border-t border-stone-100 pt-4 dark:border-stone-800">
+					<span class="text-xs font-medium tracking-wide text-stone-400 uppercase"
+						>Aktives Modell</span
+					>
+
+					{#if data.ai.connected && data.ai.models.length > 0}
+						<form
+							method="POST"
+							action="?/setModel"
+							use:enhance={() => {
+								saving = true;
+								return async ({ update }) => {
+									await update();
+									saving = false;
+								};
+							}}
+							class="mt-2 flex items-center gap-3"
+						>
+							<select
+								name="model"
+								bind:value={selectedModel}
+								class="focus:border-primary-500 focus:ring-primary-500 rounded-md border border-stone-300 bg-white px-3 py-2 text-sm text-stone-900 shadow-sm focus:ring-1 focus:outline-none dark:border-stone-600 dark:bg-stone-800 dark:text-stone-100"
+							>
+								{#each data.ai.models as model}
+									<option value={model}>{model}</option>
+								{/each}
+							</select>
+
+							<button
+								type="submit"
+								disabled={saving || selectedModel === activeModel}
+								class="bg-primary-600 hover:bg-primary-700 rounded-md px-4 py-2 text-sm font-medium text-white shadow-sm disabled:cursor-not-allowed disabled:opacity-50"
+							>
+								{saving ? 'Speichert…' : 'Übernehmen'}
+							</button>
+						</form>
+
+						{#if form?.success}
+							<p class="mt-2 text-xs text-green-600 dark:text-green-400">
+								Modell auf <span class="font-mono">{form.model}</span> gesetzt.
+							</p>
+						{/if}
+						{#if form?.error}
+							<p class="mt-2 text-xs text-red-600 dark:text-red-400">{form.error}</p>
+						{/if}
+					{:else if data.ai.connected}
+						<p class="mt-1 text-sm text-stone-500 dark:text-stone-400">
+							Keine Modelle gefunden. Installiere ein Modell mit
+							<code class="rounded bg-stone-100 px-1 py-0.5 font-mono text-xs dark:bg-stone-800">
+								ollama pull &lt;model&gt;
+							</code>
+						</p>
+					{:else}
+						<p class="mt-1 font-mono text-sm text-stone-500 dark:text-stone-400">
+							{data.ai.model || '—'}
+						</p>
+						<p class="mt-1 text-xs text-red-500">
+							Ollama nicht erreichbar — prüfe ob der Dienst läuft.
+						</p>
+					{/if}
+				</div>
+			{/if}
+		</div>
+	</div>
+</div>