feat: smart federation routing (load + VRAM + model + prefix affinity + cross-server sync) and unified Cluster UI

core/application/p2p.go

@@ -53,9 +53,21 @@ func (a *Application) StartP2P() error {
 			return err
 		}
 
+		// modelsFn reports the model names this instance currently serves so the
+		// federation proxy can route a request only to peers that have the
+		// requested model. It is re-evaluated on every announce tick.
+		modelsFn := func() []string {
+			cfgs := a.ModelConfigLoader().GetAllModelsConfigs()
+			names := make([]string, 0, len(cfgs))
+			for _, c := range cfgs {
+				names = append(names, c.Name)
+			}
+			return names
+		}
+
 		// Here a new node is created and started
 		// and a service is exposed by the node
-		node, err := p2p.ExposeService(ctx, "localhost", port, a.applicationConfig.P2PToken, p2p.NetworkID(networkID, p2p.FederatedID))
+		node, err := p2p.ExposeService(ctx, "localhost", port, a.applicationConfig.P2PToken, p2p.NetworkID(networkID, p2p.FederatedID), modelsFn)
 		if err != nil {
 			return err
 		}

core/cli/federated.go

@@ -14,12 +14,14 @@ type FederatedCLI struct {
 	RandomWorker       bool   `env:"LOCALAI_RANDOM_WORKER,RANDOM_WORKER" default:"false" help:"Select a random worker from the pool" group:"p2p"`
 	Peer2PeerNetworkID string `env:"LOCALAI_P2P_NETWORK_ID,P2P_NETWORK_ID" help:"Network ID for P2P mode, can be set arbitrarly by the user for grouping a set of instances." group:"p2p"`
 	TargetWorker       string `env:"LOCALAI_TARGET_WORKER,TARGET_WORKER" help:"Target worker to run the federated server on" group:"p2p"`
+	UploadLimit        int    `env:"LOCALAI_UPLOAD_LIMIT,UPLOAD_LIMIT" default:"15" help:"Default upload-size limit in megabytes" group:"api"`
+	AffinitySync       bool   `env:"LOCALAI_FEDERATED_AFFINITY_SYNC,FEDERATED_AFFINITY_SYNC" default:"false" help:"Broadcast prefix-cache affinity observations to other federation servers over the p2p generic channel (enable on every federation server that should cohere)" group:"p2p"`
 }
 
 func (f *FederatedCLI) Run(ctx *cliContext.Context) error {
 	warnDeprecatedFlags()
 
-	fs := p2p.NewFederatedServer(f.Address, p2p.NetworkID(f.Peer2PeerNetworkID, p2p.FederatedID), f.Peer2PeerToken, !f.RandomWorker, f.TargetWorker)
+	fs := p2p.NewFederatedServer(f.Address, p2p.NetworkID(f.Peer2PeerNetworkID, p2p.FederatedID), f.Peer2PeerToken, !f.RandomWorker, f.TargetWorker, int64(f.UploadLimit)*1024*1024, f.AffinitySync)
 
 	c, cancel := context.WithCancel(context.Background())
 

core/cli/worker/worker_p2p.go

@@ -62,7 +62,7 @@ func (r *P2P) Run(ctx *cliContext.Context) error {
 			p = r.RunnerPort
 		}
 
-		_, err = p2p.ExposeService(c, address, p, r.Token, p2p.NetworkID(r.Peer2PeerNetworkID, p2p.LlamaCPPWorkerID))
+		_, err = p2p.ExposeService(c, address, p, r.Token, p2p.NetworkID(r.Peer2PeerNetworkID, p2p.LlamaCPPWorkerID), nil)
 		if err != nil {
 			return err
 		}
@@ -104,7 +104,7 @@ func (r *P2P) Run(ctx *cliContext.Context) error {
 			}
 		}()
 
-		_, err = p2p.ExposeService(c, address, fmt.Sprint(port), r.Token, p2p.NetworkID(r.Peer2PeerNetworkID, p2p.LlamaCPPWorkerID))
+		_, err = p2p.ExposeService(c, address, fmt.Sprint(port), r.Token, p2p.NetworkID(r.Peer2PeerNetworkID, p2p.LlamaCPPWorkerID), nil)
 		if err != nil {
 			return err
 		}

core/cli/worker/worker_p2p_mlx.go

@@ -81,7 +81,7 @@ func (r *P2PMLX) Run(ctx *cliContext.Context) error {
 		}
 	}()
 
-	_, err = p2p.ExposeService(c, address, fmt.Sprint(port), r.Token, p2p.NetworkID(r.Peer2PeerNetworkID, p2p.MLXWorkerID))
+	_, err = p2p.ExposeService(c, address, fmt.Sprint(port), r.Token, p2p.NetworkID(r.Peer2PeerNetworkID, p2p.MLXWorkerID), nil)
 	if err != nil {
 		return err
 	}

core/http/react-ui/e2e/cluster.spec.js

@@ -0,0 +1,37 @@
+import { test, expect } from './coverage-fixtures.js'
+
+// The Cluster page composes two capability sections: "Distributed (NATS)" (the
+// former Nodes page) and "Swarm (p2p)" (the former P2P page). Each section only
+// mounts when its mode is enabled — distributed when /api/nodes answers OK, swarm
+// when a non-empty p2p network token is present. We mock those probes so the page
+// renders against the standalone ui-test-server without NATS / p2p running.
+
+async function mockDistributedOnly(page) {
+  await page.route('**/api/nodes', (route) => {
+    route.fulfill({ status: 200, contentType: 'application/json', body: '[]' })
+  })
+  await page.route('**/api/nodes/scheduling', (route) => {
+    route.fulfill({ status: 200, contentType: 'application/json', body: '[]' })
+  })
+  // Swarm disabled: token probe fails, so the swarm section stays hidden.
+  await page.route('**/api/p2p/token', (route) => {
+    route.fulfill({ status: 503, contentType: 'text/plain', body: '' })
+  })
+}
+
+test.describe('Cluster page', () => {
+  test('shows the page title', async ({ page }) => {
+    await mockDistributedOnly(page)
+    await page.goto('/app/cluster')
+    await expect(page).toHaveURL(/\/app\/cluster$/)
+    await expect(page.getByRole('heading', { name: /Cluster/i })).toBeVisible()
+  })
+
+  test('shows the distributed section when /api/nodes responds', async ({ page }) => {
+    await mockDistributedOnly(page)
+    await page.goto('/app/cluster')
+    await expect(page).toHaveURL(/\/app\/cluster$/)
+    // The distributed capability section is titled "Distributed (NATS)".
+    await expect(page.getByText(/Distributed \(NATS\)/i)).toBeVisible()
+  })
+})

core/http/react-ui/e2e/navigation.spec.js

@@ -23,4 +23,11 @@ test.describe('Navigation', () => {
     await expect(page).toHaveURL(/\/app\/traces/)
     await expect(page.getByRole('heading', { name: 'Traces', exact: true })).toBeVisible()
   })
+
+  test('old cluster routes redirect to /app/cluster', async ({ page }) => {
+    await page.goto('/app/p2p')
+    await expect(page).toHaveURL(/\/app\/cluster$/)
+    await page.goto('/app/nodes')
+    await expect(page).toHaveURL(/\/app\/cluster$/)
+  })
 })

core/http/react-ui/e2e/nodes-per-node-backend-actions.spec.js

@@ -81,7 +81,7 @@ async function mockDistributedNodes(page, { onDelete } = {}) {
 }
 
 async function expandNodeAndWaitForBackends(page) {
-  await page.goto('/app/nodes')
+  await page.goto('/app/cluster')
   // Click the row to expand it. The chevron toggle and the row both work,
   // but clicking the name cell is the most user-like.
   await page.getByText(NODE_NAME).first().click()

core/http/react-ui/e2e/p2p.spec.js

@@ -1,26 +1,65 @@
 import { test, expect } from './coverage-fixtures.js'
 
-// P2P (Swarm) admin page — renders in the no-auth test harness (isAdmin).
-test.describe('P2P page', () => {
-  test.beforeEach(async ({ page }) => {
-    await page.goto('/app/p2p')
+// The standalone P2P (Swarm) page was merged into the Cluster page: /app/p2p now
+// redirects to /app/cluster, and the p2p content lives under the "Swarm (p2p)"
+// section. That section only mounts when p2p is enabled (a network token is
+// present), so we mock /api/p2p/token to return a non-empty token and assert the
+// swarm content renders under the cluster page.
+const P2P_TOKEN = 'test-network-token'
+
+async function mockSwarmEnabled(page) {
+  await page.route('**/api/p2p/token', (route) => {
+    route.fulfill({
+      status: 200,
+      contentType: 'text/plain',
+      body: P2P_TOKEN,
+    })
+  })
+  await page.route('**/api/p2p/workers', (route) => {
+    route.fulfill({ status: 200, contentType: 'application/json', body: '{"nodes":[]}' })
+  })
+  await page.route('**/api/p2p/federation', (route) => {
+    route.fulfill({ status: 200, contentType: 'application/json', body: '{"nodes":[]}' })
+  })
+  await page.route('**/api/p2p/stats', (route) => {
+    route.fulfill({
+      status: 200,
+      contentType: 'application/json',
+      body: JSON.stringify({
+        llama_cpp_workers: { online: 0, total: 0 },
+        federated: { online: 0, total: 0 },
+        mlx_workers: { online: 0, total: 0 },
+      }),
+    })
   })
+  // The cluster page also probes /api/nodes for the distributed section; keep it
+  // failing (distributed disabled) so only the swarm section renders here.
+  await page.route('**/api/nodes', (route) => {
+    route.fulfill({ status: 503, contentType: 'application/json', body: '{}' })
+  })
+}
 
-  test('renders the P2P distribution overview and capability cards', async ({ page }) => {
-    await expect(page).toHaveURL(/\/app\/p2p$/)
-    await expect(page.getByRole('heading', { name: /P2P Distribution Not Enabled/i })).toBeVisible()
-    await expect(page.getByRole('heading', { name: 'Instance Federation' })).toBeVisible()
-    await expect(page.getByRole('heading', { name: 'Model Sharding' })).toBeVisible()
-    await expect(page.getByRole('heading', { name: 'Resource Sharing' })).toBeVisible()
-    await expect(page.getByRole('heading', { name: /How to Enable P2P/i })).toBeVisible()
+test.describe('P2P (Swarm) section on the Cluster page', () => {
+  test('the old /app/p2p route lands on the cluster page', async ({ page }) => {
+    await mockSwarmEnabled(page)
+    // /app/p2p redirects to /app/cluster.
+    await page.goto('/app/p2p')
+    await expect(page).toHaveURL(/\/app\/cluster$/)
+    await expect(page.getByRole('heading', { name: /Cluster/i })).toBeVisible()
   })
 
-  test('hardware selector offers build targets and responds to selection', async ({ page }) => {
-    const cpu = page.getByRole('button').filter({ hasText: /^CPU$/ })
-    const cuda = page.getByRole('button').filter({ hasText: /^CUDA 12$/ })
-    await expect(cpu).toBeVisible()
-    await expect(cuda).toBeVisible()
-    await cuda.click() // selecting a build target must not break the page
-    await expect(page.getByRole('heading', { name: /How to Enable P2P/i })).toBeVisible()
+  test('renders the Swarm (p2p) section when p2p is enabled', async ({ page }) => {
+    await mockSwarmEnabled(page)
+    await page.goto('/app/cluster')
+    await expect(page).toHaveURL(/\/app\/cluster$/)
+
+    // The collapsible swarm section is titled "Swarm (p2p)".
+    await expect(page.getByText(/Swarm \(p2p\)/i)).toBeVisible()
+
+    // The enabled p2p content (Network Token panel + the federation / sharding
+    // tabs) is rendered inside the swarm section.
+    await expect(page.getByRole('heading', { name: /Network Token/i })).toBeVisible()
+    await expect(page.getByText('Federation', { exact: true })).toBeVisible()
+    await expect(page.getByText('Model Sharding', { exact: true })).toBeVisible()
   })
 })

core/http/react-ui/public/locales/de/admin.json

@@ -58,5 +58,21 @@
   "explorer": {
     "title": "Explorer",
     "subtitle": "Dateien und Konfiguration durchsuchen"
+  },
+  "cluster": {
+    "title": "Cluster",
+    "subtitle": "Verteilte und Peer-to-Peer-Knoten, die diese Instanz bedienen",
+    "empty": "Es ist kein verteiltes oder p2p-Clustering aktiviert. Starte LocalAI im verteilten oder föderierten/p2p-Modus, um hier Cluster-Knoten zu verwalten.",
+    "distributed": {
+      "title": "Verteilt (NATS)"
+    },
+    "swarm": {
+      "title": "Swarm (p2p)"
+    },
+    "summary": {
+      "nodes": "Verteilte Knoten",
+      "inFlight": "Laufende Anfragen",
+      "peers": "Swarm-Peers online"
+    }
   }
 }

core/http/react-ui/public/locales/de/nav.json

@@ -40,6 +40,7 @@
     "traces": "Traces",
     "nodes": "Knoten",
     "swarm": "Swarm",
+    "cluster": "Cluster",
     "system": "System",
     "settings": "Einstellungen",
     "api": "API"

core/http/react-ui/public/locales/en/admin.json

@@ -81,5 +81,21 @@
   "explorer": {
     "title": "Explorer",
     "subtitle": "Browse files and configuration"
+  },
+  "cluster": {
+    "title": "Cluster",
+    "subtitle": "Distributed and peer-to-peer nodes serving this instance",
+    "empty": "No distributed or p2p clustering is enabled. Start LocalAI in distributed or federated/p2p mode to manage cluster nodes here.",
+    "distributed": {
+      "title": "Distributed (NATS)"
+    },
+    "swarm": {
+      "title": "Swarm (p2p)"
+    },
+    "summary": {
+      "nodes": "Distributed nodes",
+      "inFlight": "In-flight requests",
+      "peers": "Swarm peers online"
+    }
   }
 }

core/http/react-ui/public/locales/en/nav.json

@@ -41,6 +41,7 @@
     "traces": "Traces",
     "nodes": "Nodes",
     "swarm": "Swarm",
+    "cluster": "Cluster",
     "system": "System",
     "settings": "Settings",
     "api": "API"

core/http/react-ui/public/locales/es/admin.json

@@ -58,5 +58,21 @@
   "explorer": {
     "title": "Explorador",
     "subtitle": "Explora archivos y configuración"
+  },
+  "cluster": {
+    "title": "Clúster",
+    "subtitle": "Nodos distribuidos y entre pares que sirven a esta instancia",
+    "empty": "No hay clustering distribuido ni p2p habilitado. Inicia LocalAI en modo distribuido o federado/p2p para gestionar aquí los nodos del clúster.",
+    "distributed": {
+      "title": "Distribuido (NATS)"
+    },
+    "swarm": {
+      "title": "Swarm (p2p)"
+    },
+    "summary": {
+      "nodes": "Nodos distribuidos",
+      "inFlight": "Solicitudes en curso",
+      "peers": "Pares de Swarm en línea"
+    }
   }
 }

core/http/react-ui/public/locales/es/nav.json

@@ -40,6 +40,7 @@
     "traces": "Trazas",
     "nodes": "Nodos",
     "swarm": "Swarm",
+    "cluster": "Clúster",
     "system": "Sistema",
     "settings": "Configuración",
     "api": "API"

core/http/react-ui/public/locales/it/admin.json

@@ -58,5 +58,21 @@
   "explorer": {
     "title": "Esplora risorse",
     "subtitle": "Sfoglia file e configurazioni"
+  },
+  "cluster": {
+    "title": "Cluster",
+    "subtitle": "Nodi distribuiti e peer-to-peer al servizio di questa istanza",
+    "empty": "Nessun clustering distribuito o p2p è abilitato. Avvia LocalAI in modalità distribuita o federata/p2p per gestire qui i nodi del cluster.",
+    "distributed": {
+      "title": "Distribuito (NATS)"
+    },
+    "swarm": {
+      "title": "Swarm (p2p)"
+    },
+    "summary": {
+      "nodes": "Nodi distribuiti",
+      "inFlight": "Richieste in corso",
+      "peers": "Peer Swarm online"
+    }
   }
 }

core/http/react-ui/public/locales/it/nav.json

@@ -40,6 +40,7 @@
     "traces": "Tracce",
     "nodes": "Nodi",
     "swarm": "Swarm",
+    "cluster": "Cluster",
     "system": "Sistema",
     "settings": "Impostazioni",
     "api": "API"

core/http/react-ui/public/locales/zh-CN/admin.json

@@ -58,5 +58,21 @@
   "explorer": {
     "title": "资源浏览器",
     "subtitle": "浏览文件和配置"
+  },
+  "cluster": {
+    "title": "集群",
+    "subtitle": "为此实例提供服务的分布式和点对点节点",
+    "empty": "未启用分布式或 p2p 集群。请以分布式或联邦/p2p 模式启动 LocalAI，以便在此管理集群节点。",
+    "distributed": {
+      "title": "分布式 (NATS)"
+    },
+    "swarm": {
+      "title": "Swarm (p2p)"
+    },
+    "summary": {
+      "nodes": "分布式节点",
+      "inFlight": "进行中的请求",
+      "peers": "在线 Swarm 节点"
+    }
   }
 }

core/http/react-ui/public/locales/zh-CN/nav.json

@@ -40,6 +40,7 @@
     "traces": "追踪",
     "nodes": "节点",
     "swarm": "Swarm",
+    "cluster": "集群",
     "system": "系统",
     "settings": "设置",
     "api": "API"

core/http/react-ui/src/components/ClusterSection.jsx

@@ -0,0 +1,27 @@
+import { useState } from 'react'
+
+// ClusterSection is a collapsible, titled container for one capability area of
+// the Cluster page (Distributed / Swarm). Default expanded.
+export default function ClusterSection({ icon, title, subtitle, defaultOpen = true, children }) {
+  const [open, setOpen] = useState(defaultOpen)
+  return (
+    <section className="card" style={{ marginBottom: 'var(--spacing-lg)' }}>
+      <button
+        type="button"
+        aria-expanded={open}
+        onClick={() => setOpen((o) => !o)}
+        style={{
+          display: 'flex', alignItems: 'center', gap: 'var(--spacing-sm)',
+          width: '100%', padding: 'var(--spacing-md)', background: 'none',
+          border: 'none', cursor: 'pointer', textAlign: 'left', color: 'inherit',
+        }}
+      >
+        <i className={`fas fa-chevron-${open ? 'down' : 'right'}`} style={{ width: '1rem', color: 'var(--color-text-muted)' }} />
+        {icon && <i className={icon} style={{ color: 'var(--color-primary)' }} />}
+        <span style={{ fontWeight: 600 }}>{title}</span>
+        {subtitle && <span style={{ marginLeft: 'auto', color: 'var(--color-text-muted)', fontSize: '0.875rem' }}>{subtitle}</span>}
+      </button>
+      {open && <div style={{ padding: '0 var(--spacing-md) var(--spacing-md)' }}>{children}</div>}
+    </section>
+  )
+}