diff --git a/nemoclaw-blueprint/policies/presets/local-inference.yaml b/nemoclaw-blueprint/policies/presets/local-inference.yaml
new file mode 100644
index 0000000000..c692cdce25
--- /dev/null
+++ b/nemoclaw-blueprint/policies/presets/local-inference.yaml
@@ -0,0 +1,28 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+
+preset:
+  name: local-inference
+  description: "Local inference access (Ollama, vLLM) via host gateway"
+
+network_policies:
+  local_inference:
+    name: local_inference
+    endpoints:
+      - host: host.openshell.internal
+        port: 11434
+        protocol: rest
+        enforcement: enforce
+        rules:
+          - allow: { method: GET, path: "/**" }
+          - allow: { method: POST, path: "/**" }
+      - host: host.openshell.internal
+        port: 8000
+        protocol: rest
+        enforcement: enforce
+        rules:
+          - allow: { method: GET, path: "/**" }
+          - allow: { method: POST, path: "/**" }
+    binaries:
+      - { path: /usr/local/bin/openclaw }
+      - { path: /usr/local/bin/claude }
diff --git a/src/lib/onboard.ts b/src/lib/onboard.ts
index 4ae3f67469..0f41bda83c 100644
--- a/src/lib/onboard.ts
+++ b/src/lib/onboard.ts
@@ -46,6 +46,11 @@ const {
   getProviderSelectionConfig,
   parseGatewayInference,
 } = require("./inference-config");
+
+// Providers that run on the host and need the local-inference policy preset.
+// Shared constant so getSuggestedPolicyPresets() and setupPoliciesWithSelection()
+// stay in sync.
+const LOCAL_INFERENCE_PROVIDERS = ["ollama-local", "vllm-local"];
 const { inferContainerRuntime, isWsl, shouldPatchCoredns } = require("./platform");
 const { resolveOpenshell } = require("./resolve-openshell");
 const {
@@ -4502,8 +4507,13 @@ async function setupMessagingChannels() {
   return selected;
 }
 
-function getSuggestedPolicyPresets({ enabledChannels = null, webSearchConfig = null } = {}) {
+function getSuggestedPolicyPresets({ enabledChannels = null, webSearchConfig = null, provider = null } = {}) {
   const suggestions = ["pypi", "npm"];
+
+  // Auto-suggest local-inference preset when a local provider is selected
+  if (provider && LOCAL_INFERENCE_PROVIDERS.includes(provider)) {
+    suggestions.push("local-inference");
+  }
   const usesExplicitMessagingSelection = Array.isArray(enabledChannels);
 
   const maybeSuggestMessagingPreset = (channel, envKey) => {
@@ -5118,6 +5128,7 @@ async function setupPoliciesWithSelection(sandboxName, options = {}) {
   const onSelection = typeof options.onSelection === "function" ? options.onSelection : null;
   const webSearchConfig = options.webSearchConfig || null;
   const enabledChannels = Array.isArray(options.enabledChannels) ? options.enabledChannels : null;
+  const provider = options.provider || null;
 
   step(8, 8, "Policy presets");
 
@@ -5147,6 +5158,10 @@ async function setupPoliciesWithSelection(sandboxName, options = {}) {
   const suggestions = tiers.resolveTierPresets(tierName).map((p) => p.name);
   // Allow credential-based overrides on top of the tier (additive only).
   if (webSearchConfig && !suggestions.includes("brave")) suggestions.push("brave");
+  // Auto-suggest local-inference preset when a local provider is selected
+  if (provider && LOCAL_INFERENCE_PROVIDERS.includes(provider) && !suggestions.includes("local-inference")) {
+    suggestions.push("local-inference");
+  }
 
   if (isNonInteractive()) {
     const policyMode = (process.env.NEMOCLAW_POLICY_MODE || "suggested").trim().toLowerCase();
@@ -5966,6 +5981,7 @@ async function onboard(opts = {}) {
               : null,
           enabledChannels: selectedMessagingChannels,
           webSearchConfig,
+          provider,
           onSelection: (policyPresets) => {
             onboardSession.updateSession((current) => {
               current.policyPresets = policyPresets;
@@ -6049,6 +6065,7 @@ module.exports = {
   isOpenclawReady,
   arePolicyPresetsApplied,
   getSuggestedPolicyPresets,
+  LOCAL_INFERENCE_PROVIDERS,
   presetsCheckboxSelector,
   selectPolicyTier,
   selectTierPresetsAndAccess,
diff --git a/test/onboard.test.ts b/test/onboard.test.ts
index 35951cae3f..f1d61fa18d 100644
--- a/test/onboard.test.ts
+++ b/test/onboard.test.ts
@@ -142,6 +142,25 @@ describe("onboard helpers", () => {
     }
   });
 
+  it("suggests local-inference preset when provider is ollama-local", () => {
+    const presets = getSuggestedPolicyPresets({ provider: "ollama-local" });
+    expect(presets).toContain("local-inference");
+    expect(presets).toContain("pypi");
+    expect(presets).toContain("npm");
+  });
+
+  it("suggests local-inference preset when provider is vllm-local", () => {
+    const presets = getSuggestedPolicyPresets({ provider: "vllm-local" });
+    expect(presets).toContain("local-inference");
+  });
+
+  it("does not suggest local-inference for cloud providers", () => {
+    expect(getSuggestedPolicyPresets({ provider: "nvidia-prod" })).not.toContain("local-inference");
+    expect(getSuggestedPolicyPresets({ provider: "openai-api" })).not.toContain("local-inference");
+    expect(getSuggestedPolicyPresets({ provider: null })).not.toContain("local-inference");
+    expect(getSuggestedPolicyPresets({})).not.toContain("local-inference");
+  });
+
   it("patches the staged Dockerfile with the selected model and chat UI URL", () => {
     const tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), "nemoclaw-onboard-dockerfile-"));
     const dockerfilePath = path.join(tmpDir, "Dockerfile");
diff --git a/test/policies.test.ts b/test/policies.test.ts
index 137f94cbfe..a765e6d955 100644
--- a/test/policies.test.ts
+++ b/test/policies.test.ts
@@ -96,9 +96,9 @@ selectFromList(items, options)
 
 describe("policies", () => {
   describe("listPresets", () => {
-    it("returns all 11 presets", () => {
+    it("returns all 12 presets", () => {
       const presets = policies.listPresets();
-      expect(presets.length).toBe(11);
+      expect(presets.length).toBe(12);
     });
 
     it("each preset has name and description", () => {
@@ -120,6 +120,7 @@ describe("policies", () => {
         "github",
         "huggingface",
         "jira",
+        "local-inference",
         "npm",
         "outlook",
         "pypi",
@@ -153,6 +154,21 @@ describe("policies", () => {
         expect(content).toContain("/usr/bin/node");
       }
     });
+
+    it("local-inference preset targets host.openshell.internal on Ollama and vLLM ports", () => {
+      const content = policies.loadPreset("local-inference");
+      expect(content).toContain("host.openshell.internal");
+      expect(content).toContain("port: 11434");
+      expect(content).toContain("port: 8000");
+    });
+
+    it("local-inference preset restricts binaries to openclaw and claude", () => {
+      const content = policies.loadPreset("local-inference");
+      expect(content).toContain("/usr/local/bin/openclaw");
+      expect(content).toContain("/usr/local/bin/claude");
+      // Should NOT include node — only agent binaries need inference access
+      expect(content).not.toContain("/usr/local/bin/node");
+    });
   });
 
   describe("getPresetEndpoints", () => {