diff --git a/src/agents/pi-embedded-runner/run/attempt.ts b/src/agents/pi-embedded-runner/run/attempt.ts
index 20a33f83bcc..8c8456ced00 100644
--- a/src/agents/pi-embedded-runner/run/attempt.ts
+++ b/src/agents/pi-embedded-runner/run/attempt.ts
@@ -2141,6 +2141,7 @@ export async function runEmbeddedAttempt(
         trigger: params.trigger,
         runTimeoutMs: params.timeoutMs !== configuredRunTimeoutMs ? params.timeoutMs : undefined,
         modelRequestTimeoutMs: (params.model as { requestTimeoutMs?: number }).requestTimeoutMs,
+        model: params.model as { baseUrl?: string },
       });
       if (idleTimeoutMs > 0) {
         activeSession.agent.streamFn = streamWithIdleTimeout(
diff --git a/src/agents/pi-embedded-runner/run/llm-idle-timeout.test.ts b/src/agents/pi-embedded-runner/run/llm-idle-timeout.test.ts
index a750667dc8d..230897e4267 100644
--- a/src/agents/pi-embedded-runner/run/llm-idle-timeout.test.ts
+++ b/src/agents/pi-embedded-runner/run/llm-idle-timeout.test.ts
@@ -85,6 +85,51 @@ describe("resolveLlmIdleTimeoutMs", () => {
     const cfg = { agents: { defaults: { timeoutSeconds: 300 } } } as OpenClawConfig;
     expect(resolveLlmIdleTimeoutMs({ cfg, trigger: "cron" })).toBe(DEFAULT_LLM_IDLE_TIMEOUT_MS);
   });
+
+  it.each([
+    "http://localhost:11434",
+    "http://127.0.0.1:11434",
+    "http://0.0.0.0:11434",
+    "http://[::1]:11434",
+    "http://my-rig.local:11434",
+    "http://10.0.0.5:11434",
+    "http://172.16.5.10:11434",
+    "http://192.168.1.20:11434",
+  ])("disables the default idle watchdog for local provider baseUrl %s", (baseUrl) => {
+    expect(resolveLlmIdleTimeoutMs({ model: { baseUrl } })).toBe(0);
+  });
+
+  it("keeps the default idle watchdog for remote provider baseUrls", () => {
+    expect(resolveLlmIdleTimeoutMs({ model: { baseUrl: "https://api.openai.com/v1" } })).toBe(
+      DEFAULT_LLM_IDLE_TIMEOUT_MS,
+    );
+    expect(resolveLlmIdleTimeoutMs({ model: { baseUrl: "https://ollama.com" } })).toBe(
+      DEFAULT_LLM_IDLE_TIMEOUT_MS,
+    );
+  });
+
+  it("ignores malformed baseUrl and keeps the default idle watchdog", () => {
+    expect(resolveLlmIdleTimeoutMs({ model: { baseUrl: "not-a-url" } })).toBe(
+      DEFAULT_LLM_IDLE_TIMEOUT_MS,
+    );
+    expect(resolveLlmIdleTimeoutMs({ model: { baseUrl: "" } })).toBe(DEFAULT_LLM_IDLE_TIMEOUT_MS);
+  });
+
+  it("still honors an explicit provider request timeout for local providers", () => {
+    expect(
+      resolveLlmIdleTimeoutMs({
+        model: { baseUrl: "http://127.0.0.1:11434" },
+        modelRequestTimeoutMs: 600_000,
+      }),
+    ).toBe(600_000);
+  });
+
+  it("still applies agents.defaults.timeoutSeconds cap for local providers", () => {
+    const cfg = { agents: { defaults: { timeoutSeconds: 30 } } } as OpenClawConfig;
+    expect(resolveLlmIdleTimeoutMs({ cfg, model: { baseUrl: "http://127.0.0.1:11434" } })).toBe(
+      30_000,
+    );
+  });
 });
 
 describe("streamWithIdleTimeout", () => {
diff --git a/src/agents/pi-embedded-runner/run/llm-idle-timeout.ts b/src/agents/pi-embedded-runner/run/llm-idle-timeout.ts
index 2d680247894..cbc1e66ee08 100644
--- a/src/agents/pi-embedded-runner/run/llm-idle-timeout.ts
+++ b/src/agents/pi-embedded-runner/run/llm-idle-timeout.ts
@@ -15,6 +15,43 @@ export const DEFAULT_LLM_IDLE_TIMEOUT_MS = DEFAULT_LLM_IDLE_TIMEOUT_SECONDS * 10
  */
 const MAX_SAFE_TIMEOUT_MS = 2_147_000_000;
 
+/**
+ * Detects loopback / private-network / `.local` base URLs. Local providers
+ * (Ollama, LM Studio, llama.cpp) legitimately stay silent for many minutes
+ * during prompt evaluation and thinking, so the network-silence-as-hang
+ * heuristic that motivates the default idle watchdog does not apply.
+ */
+function isLocalProviderBaseUrl(baseUrl: string): boolean {
+  let host: string;
+  try {
+    host = new URL(baseUrl).hostname.toLowerCase();
+  } catch {
+    return false;
+  }
+  if (host.startsWith("[") && host.endsWith("]")) {
+    host = host.slice(1, -1);
+  }
+  if (
+    host === "localhost" ||
+    host === "127.0.0.1" ||
+    host === "0.0.0.0" ||
+    host === "::1" ||
+    host === "::ffff:7f00:1" ||
+    host === "::ffff:127.0.0.1" ||
+    host.endsWith(".local")
+  ) {
+    return true;
+  }
+  const octets = host.split(".").map((part) => Number.parseInt(part, 10));
+  if (octets.length !== 4 || octets.some((p) => !Number.isInteger(p) || p < 0 || p > 255)) {
+    return false;
+  }
+  const [a, b] = octets;
+  return (
+    a === 10 || (a === 172 && b !== undefined && b >= 16 && b <= 31) || (a === 192 && b === 168)
+  );
+}
+
 /**
  * Resolves the LLM idle timeout from configuration.
  * @returns Idle timeout in milliseconds, or 0 to disable
@@ -24,6 +61,7 @@ export function resolveLlmIdleTimeoutMs(params?: {
   trigger?: EmbeddedRunTrigger;
   runTimeoutMs?: number;
   modelRequestTimeoutMs?: number;
+  model?: { baseUrl?: string };
 }): number {
   const clampTimeoutMs = (valueMs: number) => Math.min(Math.floor(valueMs), MAX_SAFE_TIMEOUT_MS);
   const clampImplicitTimeoutMs = (valueMs: number) =>
@@ -72,6 +110,16 @@ export function resolveLlmIdleTimeoutMs(params?: {
     return 0;
   }
 
+  // The default watchdog is a network-silence-as-hang guard for cloud providers.
+  // Local providers can legitimately stream nothing for many minutes during
+  // prompt evaluation or thinking, so falling back to the default would abort
+  // valid local runs. Honor it only when the user has not opted out via the
+  // baseUrl pointing at loopback / private-network / `.local`.
+  const baseUrl = params?.model?.baseUrl;
+  if (typeof baseUrl === "string" && baseUrl.length > 0 && isLocalProviderBaseUrl(baseUrl)) {
+    return 0;
+  }
+
   return DEFAULT_LLM_IDLE_TIMEOUT_MS;
 }