diff --git a/CHANGELOG.md b/CHANGELOG.md
index c4e051d2..ad1ac231 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -8,6 +8,7 @@
 
 ### 修复
 
+- 模型页注册的本机 Ollama / GGUF 现在会出现在对话区模型选择器里，并走本机 `127.0.0.1:11434` 调用。注册会补齐 Ollama 供应商默认值、打开服务开关、热加载对话列表；Ollama 未启动时给出可操作错误，而不是静默失败。
 - 对话模型选择器底部增加「模型管理」，跳到侧栏同一条「模型」路由（`/models`）。
 - 组织页默认只显示本机 openXYOS 嵌入和顶部蓝色状态栏；「启用组织模块」「组织OS管理」在状态栏右下。互生长导入/导出、源码下载、循环时间线等收到「组织OS管理」。iframe 高度等于可见预览区，openXYOS 弹窗按 iframe 视口水平垂直居中。
 - 嵌入本机 openXYOS 登录不再因 CORS 漏掉 `http://127.0.0.1:3780` 而返回笼统的「服务器内部错误」。启动环境合并 sidecar 自源；无演示种子时写入本地 demo 账号，保证隔空第一台电脑能登录。
diff --git a/dashboard/src/api/modules/localModels.ts b/dashboard/src/api/modules/localModels.ts
index e24b12c2..c6b72bf2 100644
--- a/dashboard/src/api/modules/localModels.ts
+++ b/dashboard/src/api/modules/localModels.ts
@@ -58,6 +58,7 @@ export interface LocalRuntimeResult {
   size?: number;
   source?: string;
   provider_name?: string;
+  registered?: boolean;
 }
 
 export interface LocalScanJob {
diff --git a/dashboard/src/locales/en.json b/dashboard/src/locales/en.json
index caf52f49..2143c3ce 100644
--- a/dashboard/src/locales/en.json
+++ b/dashboard/src/locales/en.json
@@ -1264,7 +1264,8 @@
     "recursion_limit": "The agent hit its max iteration / recursion limit before finishing. Open Configuration and raise Max Iterations, then try again.",
     "timeout_network": "Connecting to the model timed out or the network failed. Check your network and retry. If you use a proxy or self-hosted endpoint, confirm it is reachable.",
     "provider_unavailable": "The model provider is temporarily unavailable. Retry later, or switch to another model.",
-    "model_call_failed": "The model call failed after several retries. Wait a moment and retry. If it keeps failing, check the model settings or switch models."
+    "model_call_failed": "The model call failed after several retries. Wait a moment and retry. If it keeps failing, check the model settings or switch models.",
+    "local_runtime": "The local Ollama runtime is not reachable. Start Ollama from Settings → Models, then retry. If it is already running, confirm it is listening on port 11434."
   },
   "chat": {
     "sharedExpert": {
diff --git a/dashboard/src/locales/zh.json b/dashboard/src/locales/zh.json
index 9cbe109e..ecd480ca 100644
--- a/dashboard/src/locales/zh.json
+++ b/dashboard/src/locales/zh.json
@@ -1264,7 +1264,8 @@
     "recursion_limit": "智能体已达到最大迭代次数（递归上限），任务尚未完成。请前往「运行配置」调高「最大迭代次数」后重试。",
     "timeout_network": "连接模型服务超时或网络异常。请检查网络后重试；若使用代理或自建服务，请确认其可达。",
     "provider_unavailable": "模型服务暂时不可用。请稍后重试，或切换其他模型。",
-    "model_call_failed": "模型调用多次重试后仍失败。请稍后重试；若持续失败，请检查模型配置或更换模型。"
+    "model_call_failed": "模型调用多次重试后仍失败。请稍后重试；若持续失败，请检查模型配置或更换模型。",
+    "local_runtime": "本机 Ollama 未就绪或无法连接。请到「设置 → 模型」启动 Ollama 后再试。若已启动，请确认它在 11434 端口监听。"
   },
   "chat": {
     "sharedExpert": {
diff --git a/dashboard/src/pages/Chat/hooks/useChatComposerResources.ts b/dashboard/src/pages/Chat/hooks/useChatComposerResources.ts
index 6d172ca7..c8c2c0f2 100644
--- a/dashboard/src/pages/Chat/hooks/useChatComposerResources.ts
+++ b/dashboard/src/pages/Chat/hooks/useChatComposerResources.ts
@@ -10,6 +10,7 @@ import {
 } from "../../../api/modules/knowledgeBases";
 import type { ResolvedModel } from "../../../api/types";
 import { CONNECTORS_CHANGED_EVENT } from "../../Agent/Connectors/customMcpUtils";
+import { MODELS_CHANGED_EVENT } from "../../Settings/Models/modelsChanged";
 import { useCurrentUser } from "../../../hooks/useCurrentUser";
 import { useAgent } from "../../../context/AgentContext";
 import { activeModelToRef } from "./useChatContextWindow";
@@ -273,9 +274,11 @@ export function useChatComposerResources(
     loadModels();
     const onFocus = () => loadModels();
     window.addEventListener("focus", onFocus);
+    window.addEventListener(MODELS_CHANGED_EVENT, loadModels);
     return () => {
       cancelled = true;
       window.removeEventListener("focus", onFocus);
+      window.removeEventListener(MODELS_CHANGED_EVENT, loadModels);
     };
   }, []);
 
diff --git a/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.test.tsx b/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.test.tsx
index 18314076..b6d3c834 100644
--- a/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.test.tsx
+++ b/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.test.tsx
@@ -1,6 +1,7 @@
 import { describe, expect, it, vi, beforeEach } from "vitest";
 import { render, screen, waitFor } from "@testing-library/react";
 import userEvent from "@testing-library/user-event";
+import { MemoryRouter } from "react-router-dom";
 
 const probe = vi.fn();
 const startOllama = vi.fn();
@@ -76,9 +77,17 @@ beforeEach(() => {
   });
 });
 
+function renderPanel(props: { onSaved?: () => void | Promise<void> } = {}) {
+  return render(
+    <MemoryRouter>
+      <LocalHardwarePanel {...props} />
+    </MemoryRouter>,
+  );
+}
+
 describe("<LocalHardwarePanel />", () => {
   it("shows Start Ollama when the binary exists but the API is down", async () => {
-    render(<LocalHardwarePanel />);
+    renderPanel();
     await waitFor(() => expect(probe).toHaveBeenCalled());
     expect(screen.getByText("models.localStartOllama")).toBeInTheDocument();
     await userEvent.click(screen.getByText("models.localStartOllama"));
@@ -86,16 +95,30 @@ describe("<LocalHardwarePanel />", () => {
   });
 
   it("lists discovered GGUF files with a register action", async () => {
-    render(<LocalHardwarePanel />);
+    renderPanel();
     await waitFor(() => expect(screen.getByText("tiny")).toBeInTheDocument());
     expect(screen.getByText("D:\\\\models\\\\tiny.gguf")).toBeInTheDocument();
     expect(screen.getByText("models.localRegister")).toBeInTheDocument();
   });
 
   it("starts a local weight search", async () => {
-    render(<LocalHardwarePanel />);
+    renderPanel();
     await waitFor(() => expect(probe).toHaveBeenCalled());
     await userEvent.click(screen.getByText("models.localSearchModels"));
     await waitFor(() => expect(startScan).toHaveBeenCalled());
   });
+
+  it("notifies the chat picker after a successful register", async () => {
+    const onSaved = vi.fn();
+    const heard = vi.fn();
+    register.mockResolvedValue({ ok: true, name: "tiny" });
+    window.addEventListener("octop:models-changed", heard);
+    renderPanel({ onSaved });
+    await waitFor(() => expect(screen.getByText("tiny")).toBeInTheDocument());
+    await userEvent.click(screen.getByText("models.localRegister"));
+    await waitFor(() => expect(register).toHaveBeenCalled());
+    expect(heard).toHaveBeenCalled();
+    expect(onSaved).toHaveBeenCalled();
+    window.removeEventListener("octop:models-changed", heard);
+  });
 });
diff --git a/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.tsx b/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.tsx
index a430d65a..ddf40cfd 100644
--- a/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.tsx
+++ b/dashboard/src/pages/Settings/Models/components/LocalHardwarePanel.tsx
@@ -26,32 +26,10 @@ import {
   canPickDesktopFolder,
   pickDesktopFolder,
 } from "../../../../utils/desktopFolder";
-import { providerApi } from "../../../../api/modules/provider";
 import { message } from "../../../../utils/antdMessage";
 import { apiErrorMessage } from "../../../../utils/apiError";
 import { CONVERSATION_LIST_PATH } from "../../../../layouts/conversationHome";
-import { isOllamaProviderRow } from "../presetUtils";
-
-async function registerOllamaModel(name: string): Promise<boolean> {
-  const providers = await providerApi.listProviders();
-  const ollama = providers.find(
-    (row) =>
-      row.id === "ollama" ||
-      isOllamaProviderRow({
-        name: row.name,
-        base_url: row.current_base_url,
-        api_key: row.current_api_key,
-      }),
-  );
-  if (!ollama) return false;
-  const already = [
-    ...(ollama.models ?? []),
-    ...(ollama.extra_models ?? []),
-  ].some((model) => model.id === name || model.name === name);
-  if (already) return true;
-  await providerApi.addModel(ollama.id, { id: name, name });
-  return true;
-}
+import { notifyModelsChanged } from "../modelsChanged";
 
 function mergeModels(
   ...groups: Array<LocalInstalledModel[] | undefined>
@@ -77,7 +55,11 @@ function canRegister(item: LocalInstalledModel): boolean {
   );
 }
 
-export function LocalHardwarePanel() {
+export function LocalHardwarePanel({
+  onSaved,
+}: {
+  onSaved?: () => void | Promise<void>;
+} = {}) {
   const { t } = useTranslation();
   const navigate = useNavigate();
   const [probe, setProbe] = useState<LocalProbe | null>(null);
@@ -199,18 +181,14 @@ export function LocalHardwarePanel() {
         await refresh();
         return;
       }
-      let registered = false;
-      try {
-        registered = await registerOllamaModel(name);
-      } catch {
-        registered = false;
-      }
+      notifyModelsChanged();
+      await onSaved?.();
       message.success(
         t(
-          registered
-            ? "models.localInstallRegistered"
-            : "models.localInstallNeedProvider",
-          { name },
+          result.registered === false
+            ? "models.localInstallNeedProvider"
+            : "models.localInstallRegistered",
+          { name: result.name || name },
         ),
       );
       await refresh();
@@ -338,6 +316,8 @@ export function LocalHardwarePanel() {
         showRuntimeError(result, t("models.localRegisterFailed"));
         return;
       }
+      notifyModelsChanged();
+      await onSaved?.();
       message.success(t("models.localRegisterDone", { name: result.name }));
       await refresh();
     } catch (err) {
diff --git a/dashboard/src/pages/Settings/Models/components/cards/LocalServiceCard.tsx b/dashboard/src/pages/Settings/Models/components/cards/LocalServiceCard.tsx
index 378b150e..5b07ab96 100644
--- a/dashboard/src/pages/Settings/Models/components/cards/LocalServiceCard.tsx
+++ b/dashboard/src/pages/Settings/Models/components/cards/LocalServiceCard.tsx
@@ -21,6 +21,7 @@ import type {
   ProviderPreset,
   ProviderRow,
 } from "../../useProviders";
+import { MODELS_CHANGED_EVENT } from "../../modelsChanged";
 import { presetLogoId } from "../../presetUtils";
 import { ProviderConfigModal } from "../modals/ProviderConfigModal";
 import styles from "../../index.module.less";
@@ -71,36 +72,41 @@ export function LocalServiceCard({
 
   useEffect(() => {
     let cancelled = false;
-    void (async () => {
-      try {
-        if (isOnnx) {
-          const st = await onnxModelApi.getStatus();
-          if (!cancelled) {
-            setServiceEnabled(st.enabled);
-            setServiceRunning(st.ready || st.enabled);
-            setDepsAvailable(st.deps_available !== false);
-            setDepsInstallFailed(false);
+    const refreshService = () => {
+      void (async () => {
+        try {
+          if (isOnnx) {
+            const st = await onnxModelApi.getStatus();
+            if (!cancelled) {
+              setServiceEnabled(st.enabled);
+              setServiceRunning(st.ready || st.enabled);
+              setDepsAvailable(st.deps_available !== false);
+              setDepsInstallFailed(false);
+            }
+          } else {
+            const st = await ollamaModelApi.getService();
+            if (!cancelled) {
+              setServiceEnabled(st.enabled);
+              setServiceRunning(st.running);
+              setDepsAvailable(true);
+              setDepsInstallFailed(false);
+            }
           }
-        } else {
-          const st = await ollamaModelApi.getService();
+        } catch {
           if (!cancelled) {
-            setServiceEnabled(st.enabled);
-            setServiceRunning(st.running);
-            setDepsAvailable(true);
-            setDepsInstallFailed(false);
+            setServiceEnabled(false);
+            setServiceRunning(false);
           }
         }
-      } catch {
-        if (!cancelled) {
-          setServiceEnabled(false);
-          setServiceRunning(false);
-        }
-      }
-    })();
+      })();
+    };
+    refreshService();
+    window.addEventListener(MODELS_CHANGED_EVENT, refreshService);
     return () => {
       cancelled = true;
+      window.removeEventListener(MODELS_CHANGED_EVENT, refreshService);
     };
-  }, [isOnnx, provider?.id]);
+  }, [isOnnx, provider?.id, provider?.enabled]);
 
   const ensureProvider = async (): Promise<ProviderRow> => {
     if (row) return row;
diff --git a/dashboard/src/pages/Settings/Models/index.tsx b/dashboard/src/pages/Settings/Models/index.tsx
index 41d86d1e..9abbc945 100644
--- a/dashboard/src/pages/Settings/Models/index.tsx
+++ b/dashboard/src/pages/Settings/Models/index.tsx
@@ -276,7 +276,7 @@ export default function ModelsPage() {
 
           <Divider style={{ margin: "24px 0" }} />
 
-          {showLocalTab ? <LocalHardwarePanel /> : null}
+          {showLocalTab ? <LocalHardwarePanel onSaved={fetchAll} /> : null}
 
           {showPresetSection && (
             <>
diff --git a/dashboard/src/pages/Settings/Models/modelsChanged.ts b/dashboard/src/pages/Settings/Models/modelsChanged.ts
new file mode 100644
index 00000000..bf5336e3
--- /dev/null
+++ b/dashboard/src/pages/Settings/Models/modelsChanged.ts
@@ -0,0 +1,6 @@
+export const MODELS_CHANGED_EVENT = "octop:models-changed";
+
+export function notifyModelsChanged(): void {
+  if (typeof window === "undefined") return;
+  window.dispatchEvent(new CustomEvent(MODELS_CHANGED_EVENT));
+}
diff --git a/dashboard/src/pages/Settings/Models/presetUtils.test.ts b/dashboard/src/pages/Settings/Models/presetUtils.test.ts
new file mode 100644
index 00000000..e2e4d98c
--- /dev/null
+++ b/dashboard/src/pages/Settings/Models/presetUtils.test.ts
@@ -0,0 +1,41 @@
+import { describe, expect, it } from "vitest";
+import { findConfiguredProvider } from "./presetUtils";
+import type { ProviderPreset, ProviderRow } from "./useProviders";
+
+const ollamaPreset = {
+  id: "ollama",
+  name: "Ollama (Local)",
+} as ProviderPreset;
+
+function row(
+  partial: Partial<ProviderRow> & Pick<ProviderRow, "name">,
+): ProviderRow {
+  return {
+    id: 1,
+    kind: "openai",
+    base_url: null,
+    api_key: null,
+    models: [],
+    note: null,
+    enabled: true,
+    ...partial,
+  };
+}
+
+describe("findConfiguredProvider", () => {
+  it("matches the Ollama preset by display name or id", () => {
+    const named = row({ name: "Ollama (Local)" });
+    expect(findConfiguredProvider(ollamaPreset, [named])).toEqual(named);
+    const short = row({ name: "ollama" });
+    expect(findConfiguredProvider(ollamaPreset, [short])).toEqual(short);
+  });
+
+  it("matches a local Ollama row that only shares the 11434 endpoint", () => {
+    const local = row({
+      name: "Desktop Llama",
+      base_url: "http://127.0.0.1:11434/v1",
+      api_key: "ollama",
+    });
+    expect(findConfiguredProvider(ollamaPreset, [local])).toEqual(local);
+  });
+});
diff --git a/dashboard/src/pages/Settings/Models/presetUtils.ts b/dashboard/src/pages/Settings/Models/presetUtils.ts
index fdf7ca48..359b62be 100644
--- a/dashboard/src/pages/Settings/Models/presetUtils.ts
+++ b/dashboard/src/pages/Settings/Models/presetUtils.ts
@@ -65,13 +65,6 @@ export function presetLogoId(preset: ProviderPreset): string {
   return preset.logo_id || preset.id;
 }
 
-export function findConfiguredProvider(
-  preset: ProviderPreset,
-  providers: ProviderRow[],
-): ProviderRow | undefined {
-  return providers.find((p) => p.name === preset.name || p.name === preset.id);
-}
-
 export function groupPresets(presets: ProviderPreset[]): {
   grouped: PresetGroup[];
   ungrouped: ProviderPreset[];
@@ -167,3 +160,20 @@ export function isLocalProviderRow(provider: {
 }): boolean {
   return isOnnxProviderRow(provider) || isOllamaProviderRow(provider);
 }
+
+export function findConfiguredProvider(
+  preset: ProviderPreset,
+  providers: ProviderRow[],
+): ProviderRow | undefined {
+  const exact = providers.find(
+    (p) => p.name === preset.name || p.name === preset.id,
+  );
+  if (exact) return exact;
+  if (preset.id === "ollama") {
+    return providers.find((p) => isOllamaProviderRow(p));
+  }
+  if (preset.id === "onnx") {
+    return providers.find((p) => isOnnxProviderRow(p));
+  }
+  return undefined;
+}
diff --git a/dashboard/src/utils/chatStreamError.test.ts b/dashboard/src/utils/chatStreamError.test.ts
index 2faf867c..4ca2ce53 100644
--- a/dashboard/src/utils/chatStreamError.test.ts
+++ b/dashboard/src/utils/chatStreamError.test.ts
@@ -47,6 +47,15 @@ describe("classifyChatStreamError", () => {
     });
   });
 
+  it("classifies an unreachable local Ollama runtime", () => {
+    const msg = "Failed to connect to ollama at http://127.0.0.1:11434";
+    expect(classifyChatStreamError(msg)).toBe("stream_errors.local_runtime");
+    expect(chatStreamErrorAction(msg)).toEqual({
+      path: "/models",
+      labelKey: "models.useInChat",
+    });
+  });
+
   it("classifies HTTP 5xx as provider_unavailable", () => {
     expect(classifyChatStreamError("HTTP 503: service overloaded")).toBe(
       "stream_errors.provider_unavailable",
diff --git a/dashboard/src/utils/chatStreamError.ts b/dashboard/src/utils/chatStreamError.ts
index 9773e0a2..7b5304ca 100644
--- a/dashboard/src/utils/chatStreamError.ts
+++ b/dashboard/src/utils/chatStreamError.ts
@@ -12,6 +12,7 @@ const _STREAM_ERROR_KEYS = [
   "stream_errors.timeout_network",
   "stream_errors.provider_unavailable",
   "stream_errors.model_call_failed",
+  "stream_errors.local_runtime",
 ] as const;
 
 export type StreamErrorKey = (typeof _STREAM_ERROR_KEYS)[number];
@@ -35,6 +36,10 @@ const STREAM_ERROR_ACTIONS: Partial<Record<StreamErrorKey, StreamErrorAction>> =
       path: "/system-settings/agent-config",
       labelKey: "chat.goToAgentConfig",
     },
+    "stream_errors.local_runtime": {
+      path: "/models",
+      labelKey: "models.useInChat",
+    },
   };
 
 function normalizeMessage(message: string): string {
@@ -126,6 +131,18 @@ export function classifyChatStreamError(
     return "stream_errors.recursion_limit";
   }
 
+  if (
+    (lower.includes("11434") || lower.includes("ollama")) &&
+    (lower.includes("connect") ||
+      lower.includes("refused") ||
+      lower.includes("unreachable") ||
+      lower.includes("not running") ||
+      lower.includes("connection") ||
+      lower.includes("errno 111"))
+  ) {
+    return "stream_errors.local_runtime";
+  }
+
   if (
     compact.includes("internalservererror") ||
     lower.includes("bad gateway") ||
diff --git a/src/octop/api/routers/local_models.py b/src/octop/api/routers/local_models.py
index 761d9092..16c94ce9 100644
--- a/src/octop/api/routers/local_models.py
+++ b/src/octop/api/routers/local_models.py
@@ -3,7 +3,6 @@
 from __future__ import annotations
 
 import asyncio
-import json
 import logging
 import subprocess
 from typing import Any
@@ -13,14 +12,18 @@
 
 from octop.api.deps import current_user, get_server, require_permission
 from octop.infra.agents.providers.local_probe import probe_local_models
-from octop.infra.agents.providers.local_register import load_registered, register_local_weight
+from octop.infra.agents.providers.local_register import (
+    ensure_ollama_service_flag,
+    load_registered,
+    register_local_weight,
+    upsert_ollama_model,
+)
 from octop.infra.agents.providers.local_scan import (
     cancel_scan_job,
     get_scan_job,
     latest_scan_job,
     start_scan_job,
 )
-from octop.infra.agents.providers.model_flags import is_ollama_local_provider
 from octop.infra.agents.providers.ollama_install import ensure_ollama_runtime
 from octop.infra.errors import ErrorCode, OctopError
 from octop.infra.server import OctopServer
@@ -137,37 +140,26 @@ async def local_models_ensure_deps(
     return await asyncio.to_thread(ensure_ollama_runtime, install=body.install)
 
 
-def _register_ollama_model(server: OctopServer, name: str) -> bool:
+def _register_installed_ollama(server: OctopServer, name: str) -> str:
+    """Persist *name* on the shared Ollama provider and mark the runtime on."""
     services = getattr(server, "services", None)
     repo = getattr(services, "provider_repo", None) if services is not None else None
     if repo is None:
-        return False
-    model = {"id": name, "name": name, "enabled": True, "input": ["text"]}
-    row = None
-    for candidate in repo.list_all():
-        if is_ollama_local_provider(
-            candidate.name,
-            provider_api_key=candidate.api_key,
-            provider_base_url=candidate.base_url,
-        ):
-            row = candidate
-            break
-    if row is None:
-        repo.create(
-            name="ollama",
-            kind="ollama",
-            base_url="http://127.0.0.1:11434/v1",
-            api_key="ollama",
-            models_json=json.dumps([model]),
-            note="Registered by one-click local model install",
-        )
-        return True
-    models = row.get_models()
-    if any(str(item.get("id") or "") == name for item in models if isinstance(item, dict)):
-        return False
-    models.append(model)
-    repo.update(row.id, models_json=json.dumps(models))
-    return True
+        raise OSError("provider store is not available")
+    provider_name = upsert_ollama_model(repo, name, name)
+    settings = getattr(services, "settings_repo", None)
+    if settings is not None:
+        ensure_ollama_service_flag(settings)
+    return provider_name
+
+
+async def _reload_after_register(server: OctopServer, provider_name: str, model_name: str) -> None:
+    if server.app_runtime is None:
+        return
+    try:
+        await server.app_runtime.agent_registry.on_provider_changed(provider_name=provider_name)
+    except Exception as exc:
+        logger.warning("Registered %s but provider reload failed: %s", model_name, exc)
 
 
 @router.post("/install", summary="Download and register a recommended Ollama model")
@@ -187,16 +179,19 @@ async def local_models_install(
         info = await asyncio.to_thread(OllamaModelManager.pull_model, body.name)
     except Exception as exc:
         raise register_failure(exc) from exc
+    model_name = info.name or body.name
     try:
-        registered = _register_ollama_model(server, info.name or body.name)
+        provider_name = _register_installed_ollama(server, model_name)
     except Exception as exc:
         raise register_failure(exc) from exc
+    await _reload_after_register(server, provider_name, model_name)
     return {
         "ok": True,
-        "name": info.name,
+        "name": model_name,
         "size": info.size,
         "source": "ollama",
-        "registered": registered,
+        "registered": True,
+        "provider_name": provider_name,
     }
 
 
@@ -259,15 +254,10 @@ async def local_models_register(
         )
     except _REGISTER_FAIL_TYPES as exc:
         raise register_failure(exc) from exc
-    if server.app_runtime is not None and result.get("ok"):
-        try:
-            await server.app_runtime.agent_registry.on_provider_changed(
-                provider_name=str(result.get("provider_name") or "Ollama (Local)")
-            )
-        except Exception as exc:
-            logger.warning(
-                "Registered %s but provider reload failed: %s",
-                result.get("name"),
-                exc,
-            )
+    if result.get("ok"):
+        await _reload_after_register(
+            server,
+            str(result.get("provider_name") or "Ollama (Local)"),
+            str(result.get("name") or body.name or ""),
+        )
     return result
diff --git a/src/octop/api/routers/ollama_models.py b/src/octop/api/routers/ollama_models.py
index b987a804..4d42e2c8 100644
--- a/src/octop/api/routers/ollama_models.py
+++ b/src/octop/api/routers/ollama_models.py
@@ -24,6 +24,7 @@
     get_tasks,
     update_status,
 )
+from octop.infra.agents.providers.model_flags import OLLAMA_SERVICE_SETTINGS_KEY
 
 logger = logging.getLogger(__name__)
 
@@ -233,9 +234,6 @@ async def delete_ollama_model(
     return {"status": "deleted", "name": name}
 
 
-_SETTINGS_KEY_OLLAMA_SERVICE = "ollama_service_enabled"
-
-
 class OllamaServiceBody(BaseModel):
     enabled: bool = Field(..., description="Whether Octop should keep the Ollama service running")
 
@@ -246,7 +244,7 @@ class OllamaServiceStatus(BaseModel):
 
 
 def _ollama_service_enabled(server: Any) -> bool:
-    raw = server.services.settings_repo.get(_SETTINGS_KEY_OLLAMA_SERVICE)
+    raw = server.services.settings_repo.get(OLLAMA_SERVICE_SETTINGS_KEY)
     if raw is None:
         return False
     return raw.strip().lower() in {"1", "true", "yes", "on"}
@@ -289,7 +287,7 @@ async def put_ollama_service(
     )
 
     server.services.settings_repo.set(
-        _SETTINGS_KEY_OLLAMA_SERVICE,
+        OLLAMA_SERVICE_SETTINGS_KEY,
         "true" if body.enabled else "false",
     )
     if body.enabled:
diff --git a/src/octop/i18n/domains/stream.py b/src/octop/i18n/domains/stream.py
index 05d4bb6c..4d627087 100644
--- a/src/octop/i18n/domains/stream.py
+++ b/src/octop/i18n/domains/stream.py
@@ -16,11 +16,13 @@
 TIMEOUT_NETWORK = f"{_PREFIX}stream_errors.timeout_network"
 PROVIDER_UNAVAILABLE = f"{_PREFIX}stream_errors.provider_unavailable"
 MODEL_CALL_FAILED = f"{_PREFIX}stream_errors.model_call_failed"
+LOCAL_RUNTIME = f"{_PREFIX}stream_errors.local_runtime"
 
 __all__ = [
     "AUTH",
     "CONTEXT_LENGTH",
     "INSUFFICIENT_BALANCE",
+    "LOCAL_RUNTIME",
     "MODEL_CALL_FAILED",
     "PROVIDER_UNAVAILABLE",
     "RATE_LIMIT",
@@ -118,6 +120,16 @@ def classify_stream_error_message(message: str) -> str | None:
     ):
         return RECURSION_LIMIT
 
+    if ("11434" in lower or "ollama" in lower) and (
+        "connect" in lower
+        or "refused" in lower
+        or "unreachable" in lower
+        or "not running" in lower
+        or "connection" in lower
+        or "errno 111" in lower
+    ):
+        return LOCAL_RUNTIME
+
     if (
         "internalservererror" in compact
         or "bad gateway" in lower
diff --git a/src/octop/i18n/en.json b/src/octop/i18n/en.json
index e1cadd18..c6493991 100644
--- a/src/octop/i18n/en.json
+++ b/src/octop/i18n/en.json
@@ -596,7 +596,8 @@
     "recursion_limit": "The agent hit its max iteration / recursion limit before finishing. Open Configuration and raise Max Iterations, then try again.",
     "timeout_network": "Connecting to the model timed out or the network failed. Check your network and retry. If you use a proxy or self-hosted endpoint, confirm it is reachable.",
     "provider_unavailable": "The model provider is temporarily unavailable. Retry later, or switch to another model.",
-    "model_call_failed": "The model call failed after several retries. Wait a moment and retry. If it keeps failing, check the model settings or switch models."
+    "model_call_failed": "The model call failed after several retries. Wait a moment and retry. If it keeps failing, check the model settings or switch models.",
+    "local_runtime": "The local Ollama runtime is not reachable. Start Ollama from Settings → Models, then retry. If it is already running, confirm it is listening on port 11434."
   },
   "attachment": {
     "header": "[Attachment] {filename}",
diff --git a/src/octop/i18n/zh.json b/src/octop/i18n/zh.json
index a8c5d1e7..28d9d639 100644
--- a/src/octop/i18n/zh.json
+++ b/src/octop/i18n/zh.json
@@ -596,7 +596,8 @@
     "recursion_limit": "智能体已达到最大迭代次数（递归上限），任务尚未完成。请前往「运行配置」调高「最大迭代次数」后重试。",
     "timeout_network": "连接模型服务超时或网络异常。请检查网络后重试；若使用代理或自建服务，请确认其可达。",
     "provider_unavailable": "模型服务暂时不可用。请稍后重试，或切换其他模型。",
-    "model_call_failed": "模型调用多次重试后仍失败。请稍后重试；若持续失败，请检查模型配置或更换模型。"
+    "model_call_failed": "模型调用多次重试后仍失败。请稍后重试；若持续失败，请检查模型配置或更换模型。",
+    "local_runtime": "本机 Ollama 未就绪或无法连接。请到「设置 → 模型」启动 Ollama 后再试。若已启动，请确认它在 11434 端口监听。"
   },
   "attachment": {
     "header": "[附件] {filename}",
diff --git a/src/octop/infra/agents/providers/local_register.py b/src/octop/infra/agents/providers/local_register.py
index 6cb6c9c3..034057ba 100644
--- a/src/octop/infra/agents/providers/local_register.py
+++ b/src/octop/infra/agents/providers/local_register.py
@@ -8,7 +8,12 @@
 from pathlib import Path
 from typing import Any
 
-from octop.infra.agents.providers.model_flags import is_ollama_local_provider
+from octop.infra.agents.providers.model_flags import (
+    OLLAMA_DEFAULT_BASE_URL,
+    OLLAMA_PROVIDER_DISPLAY_NAME,
+    OLLAMA_SERVICE_SETTINGS_KEY,
+    is_ollama_local_provider,
+)
 from octop.infra.utils.ollama_manager import OllamaModelManager, create_from_weight
 
 logger = logging.getLogger(__name__)
@@ -16,7 +21,6 @@
 _SETTINGS_KEY = "local_registered_weights"
 # Keep Ollama tags (``llama3.2:1b``) while sanitizing Windows paths / file stems.
 _NAME_SAFE = re.compile(r"[^a-z0-9._:-]+")
-_OLLAMA_BASE_URL = "http://127.0.0.1:11434/v1"
 _WEIGHT_SOURCES = frozenset({"gguf", "ggml"})
 _REGISTER_SOURCES = frozenset({*_WEIGHT_SOURCES, "ollama"})
 
@@ -70,7 +74,16 @@ def _find_ollama_row(provider_repo: Any) -> Any | None:
     return None
 
 
-def _upsert_ollama_model(provider_repo: Any, model_id: str, display: str) -> str:
+def ensure_ollama_service_flag(settings_repo: Any) -> None:
+    """Mark the Ollama runtime as enabled so chat / Models cards stay in sync."""
+    setter = getattr(settings_repo, "set", None)
+    if not callable(setter):
+        return
+    setter(OLLAMA_SERVICE_SETTINGS_KEY, "true")
+
+
+def upsert_ollama_model(provider_repo: Any, model_id: str, display: str) -> str:
+    """Add *model_id* to the local Ollama provider and make it chat-usable."""
     row = _find_ollama_row(provider_repo)
     model = {
         "id": model_id,
@@ -81,19 +94,19 @@ def _upsert_ollama_model(provider_repo: Any, model_id: str, display: str) -> str
     if row is None:
         try:
             provider_repo.create(
-                name="Ollama (Local)",
-                kind="ollama",
-                base_url=_OLLAMA_BASE_URL,
+                name=OLLAMA_PROVIDER_DISPLAY_NAME,
+                kind="openai",
+                base_url=OLLAMA_DEFAULT_BASE_URL,
                 api_key="ollama",
                 models_json=json.dumps([model]),
             )
-            return "Ollama (Local)"
+            return OLLAMA_PROVIDER_DISPLAY_NAME
         except Exception as exc:
             existing = getattr(provider_repo, "get_by_name", None)
-            row = existing("Ollama (Local)") if callable(existing) else None
+            row = existing(OLLAMA_PROVIDER_DISPLAY_NAME) if callable(existing) else None
             if row is None:
                 raise OSError(f"Could not save the Ollama provider: {exc}") from exc
-    models = row.get_models()
+    models = [item for item in row.get_models() if isinstance(item, dict)]
     existing = next((item for item in models if str(item.get("id") or "") == model_id), None)
     if existing is None:
         models.append(model)
@@ -104,7 +117,9 @@ def _upsert_ollama_model(provider_repo: Any, model_id: str, display: str) -> str
         row.id,
         models_json=json.dumps(models),
         enabled=True,
-        base_url=row.base_url or _OLLAMA_BASE_URL,
+        kind=getattr(row, "kind", None) or "openai",
+        api_key=getattr(row, "api_key", None) or "ollama",
+        base_url=getattr(row, "base_url", None) or OLLAMA_DEFAULT_BASE_URL,
     )
     return str(row.name)
 
@@ -188,7 +203,7 @@ def register_local_weight(
     except Exception as exc:
         logger.warning("Ollama list after register failed for %s: %s", model_id, exc)
     try:
-        provider_name = _upsert_ollama_model(provider_repo, model_id, model_id)
+        provider_name = upsert_ollama_model(provider_repo, model_id, model_id)
     except Exception as exc:
         return _fail(f"Could not save the model provider: {exc}")
     entry = {
@@ -204,6 +219,12 @@ def register_local_weight(
         remember_weight(settings_repo, entry)
     except Exception as exc:
         return _fail(f"Could not remember the registered model: {exc}")
+    try:
+        ensure_ollama_service_flag(settings_repo)
+    except Exception as exc:
+        logger.warning(
+            "Registered %s but could not enable the Ollama service flag: %s", model_id, exc
+        )
     return {
         "ok": True,
         "action": "registered",
diff --git a/src/octop/infra/agents/providers/model_flags.py b/src/octop/infra/agents/providers/model_flags.py
index e654abab..3718b490 100644
--- a/src/octop/infra/agents/providers/model_flags.py
+++ b/src/octop/infra/agents/providers/model_flags.py
@@ -11,6 +11,9 @@
 # Same pattern for Ollama (placeholder api_key + exact preset names).
 _OLLAMA_PRESET_API_KEY = "ollama"
 _OLLAMA_PRESET_NAMES = frozenset({"ollama", "ollama (local)"})
+OLLAMA_PROVIDER_DISPLAY_NAME = "Ollama (Local)"
+OLLAMA_DEFAULT_BASE_URL = "http://127.0.0.1:11434/v1"
+OLLAMA_SERVICE_SETTINGS_KEY = "ollama_service_enabled"
 
 
 def is_onnx_local_provider(
@@ -52,6 +55,24 @@ def is_ollama_local_provider(
     return "11434" in url or "ollama" in url
 
 
+def local_ollama_endpoint(
+    provider_name: str | None = None,
+    *,
+    provider_api_key: str | None = None,
+    provider_base_url: str | None = None,
+) -> tuple[str | None, str | None]:
+    """Return ``(api_key, base_url)``, filling Ollama defaults for local rows."""
+    api_key = (provider_api_key or "").strip() or None
+    base_url = (provider_base_url or "").strip() or None
+    if is_ollama_local_provider(
+        provider_name,
+        provider_api_key=api_key,
+        provider_base_url=base_url,
+    ):
+        return api_key or _OLLAMA_PRESET_API_KEY, base_url or OLLAMA_DEFAULT_BASE_URL
+    return api_key, base_url
+
+
 def is_local_runtime_provider(
     provider_name: str | None = None,
     *,
diff --git a/src/octop/infra/agents/providers/resolved.py b/src/octop/infra/agents/providers/resolved.py
index 3f09a535..8389db80 100644
--- a/src/octop/infra/agents/providers/resolved.py
+++ b/src/octop/infra/agents/providers/resolved.py
@@ -4,7 +4,10 @@
 
 from typing import Any
 
-from octop.infra.agents.providers.model_flags import is_chat_eligible_model
+from octop.infra.agents.providers.model_flags import (
+    is_chat_eligible_model,
+    local_ollama_endpoint,
+)
 from octop.infra.agents.providers.reasoning import reasoning_capability
 
 
@@ -16,14 +19,20 @@ def list_resolved_models(providers: list[Any]) -> list[dict[str, Any]]:
     """
     resolved: list[dict[str, Any]] = []
     for provider in providers:
-        if not provider.enabled or not provider.api_key:
+        if not provider.enabled:
             continue
         provider_name = str(getattr(provider, "name", "") or "")
         provider_api_key = str(getattr(provider, "api_key", "") or "") or None
+        provider_base_url = str(getattr(provider, "base_url", "") or "") or None
+        api_key, _base_url = local_ollama_endpoint(
+            provider_name,
+            provider_api_key=provider_api_key,
+            provider_base_url=provider_base_url,
+        )
+        if not api_key:
+            continue
         for m in provider.get_models():
-            if not is_chat_eligible_model(
-                m, provider_name=provider_name, provider_api_key=provider_api_key
-            ):
+            if not is_chat_eligible_model(m, provider_name=provider_name, provider_api_key=api_key):
                 continue
             # Total context drives the UI ring; prompt and output caps remain
             # separately available for runtime budgeting and model settings.
diff --git a/src/octop/infra/agents/providers/store.py b/src/octop/infra/agents/providers/store.py
index e419b241..611985e7 100644
--- a/src/octop/infra/agents/providers/store.py
+++ b/src/octop/infra/agents/providers/store.py
@@ -8,7 +8,12 @@
 
 from harness_agent.config import ModelConfig, ProviderConfig
 
-from octop.infra.agents.providers.model_flags import is_chat_eligible_model, is_vision_model
+from octop.infra.agents.providers.model_flags import (
+    is_chat_eligible_model,
+    is_ollama_local_provider,
+    is_vision_model,
+    local_ollama_endpoint,
+)
 from octop.infra.agents.providers.opencode_session import (
     OPENCODE_SESSION_HEADER,
     is_opencode_go_base_url,
@@ -111,18 +116,44 @@ def __init__(
     ) -> None:
         self._provider_repo = provider_repo
 
+    def _row_credentials(self, row: Any) -> tuple[str | None, str | None]:
+        return local_ollama_endpoint(
+            getattr(row, "name", None),
+            provider_api_key=getattr(row, "api_key", None),
+            provider_base_url=getattr(row, "base_url", None),
+        )
+
+    def _row_for_model_ref(self, ref: str) -> tuple[Any | None, str]:
+        """Resolve ``provider/model``; local aliases like ``ollama/`` map to the Ollama row."""
+        provider_name, _, model_id = ref.partition("/")
+        if not provider_name or not model_id:
+            return None, model_id
+        row = self._provider_repo.get_by_name(provider_name)
+        if row is not None:
+            return row, model_id
+        if is_ollama_local_provider(provider_name):
+            for candidate in self._provider_repo.list_all():
+                if is_ollama_local_provider(
+                    candidate.name,
+                    provider_api_key=candidate.api_key,
+                    provider_base_url=candidate.base_url,
+                ):
+                    return candidate, model_id
+        return None, model_id
+
     def iter_usable_rows(self) -> Iterator[Any]:
         """Yield enabled DB providers that have credentials and at least one enabled model."""
         for row in self._provider_repo.list_all():
             if not row.enabled:
                 continue
-            if not row.base_url or not row.api_key:
+            api_key, base_url = self._row_credentials(row)
+            if not base_url or not api_key:
                 continue
             models = row.get_models()
             if not models:
                 continue
             if any(
-                is_chat_eligible_model(m, provider_name=row.name, provider_api_key=row.api_key)
+                is_chat_eligible_model(m, provider_name=row.name, provider_api_key=api_key)
                 for m in models
             ):
                 yield row
@@ -135,11 +166,14 @@ def build_harness_configs(self) -> list[ProviderConfig]:
         out: list[ProviderConfig] = []
         for row in self.iter_usable_rows():
             protocol = KIND_TO_PROTOCOL.get(row.kind, "openai")
+            api_key, base_url = self._row_credentials(row)
+            if not api_key or not base_url:
+                continue
             raw_models = json.loads(row.models_json) if getattr(row, "models_json", None) else []
             models = [
                 self._model_config_from_row(m)
                 for m in raw_models
-                if is_chat_eligible_model(m, provider_name=row.name, provider_api_key=row.api_key)
+                if is_chat_eligible_model(m, provider_name=row.name, provider_api_key=api_key)
             ]
             if not models:
                 continue
@@ -154,14 +188,14 @@ def build_harness_configs(self) -> list[ProviderConfig]:
             out.append(
                 ProviderConfig(
                     id=row.name,
-                    base_url=row.base_url,
-                    api_key=row.api_key,
+                    base_url=base_url,
+                    api_key=api_key,
                     protocol=protocol,  # type: ignore[arg-type]
                     name=row.name,
                     models=models,
                     headers=headers,
                     session_header=(
-                        OPENCODE_SESSION_HEADER if is_opencode_go_base_url(row.base_url) else None
+                        OPENCODE_SESSION_HEADER if is_opencode_go_base_url(base_url) else None
                     ),
                 )
             )
@@ -183,13 +217,13 @@ def is_model_ref_multimodal(self, ref: str) -> bool:
         ref = ref.strip()
         if not ref or "/" not in ref:
             return False
-        provider_name, _, model_id = ref.partition("/")
-        row = self._provider_repo.get_by_name(provider_name)
-        if row is None:
+        row, model_id = self._row_for_model_ref(ref)
+        if row is None or not model_id:
             return False
+        api_key, _base_url = self._row_credentials(row)
         for model in row.get_models():
             if model.get("id") == model_id and is_chat_eligible_model(
-                model, provider_name=provider_name, provider_api_key=row.api_key
+                model, provider_name=row.name, provider_api_key=api_key
             ):
                 return _model_dict_supports_image(model)
         return False
@@ -239,27 +273,24 @@ def is_model_ref_usable(self, ref: str) -> bool:
         ref = ref.strip()
         if not ref or ref.lower() == "auto" or "/" not in ref:
             return False
-        provider_name, _, model_id = ref.partition("/")
-        if not provider_name or not model_id:
+        row, model_id = self._row_for_model_ref(ref)
+        if row is None or not row.enabled or not model_id:
             return False
-        row = self._provider_repo.get_by_name(provider_name)
-        if row is None or not row.enabled or not row.api_key or not row.base_url:
+        api_key, base_url = self._row_credentials(row)
+        if not api_key or not base_url:
             return False
         for model in row.get_models():
             if model.get("id") != model_id:
                 continue
-            return is_chat_eligible_model(
-                model, provider_name=provider_name, provider_api_key=row.api_key
-            )
+            return is_chat_eligible_model(model, provider_name=row.name, provider_api_key=api_key)
         return False
 
     def get_model_reasoning_capability(self, ref: str) -> dict[str, Any] | None:
         """Return normalized reasoning metadata for a usable model ref."""
         if not self.is_model_ref_usable(ref):
             return None
-        provider_name, _, model_id = ref.partition("/")
-        row = self._provider_repo.get_by_name(provider_name)
-        if row is None:
+        row, model_id = self._row_for_model_ref(ref)
+        if row is None or not model_id:
             return None
         for model in row.get_models():
             if model.get("id") == model_id and model.get("enabled", True):
diff --git a/tests/unit/agents/test_model_flags.py b/tests/unit/agents/test_model_flags.py
index fa497566..903d9f69 100644
--- a/tests/unit/agents/test_model_flags.py
+++ b/tests/unit/agents/test_model_flags.py
@@ -127,3 +127,19 @@ def test_list_resolved_models_skips_embedding() -> None:
     assert resolved[0]["max_input_tokens"] == 120_000
     assert resolved[0]["max_output_tokens"] == 16_384
     assert resolved[0]["max_tokens"] == 16_384
+
+
+def test_list_resolved_models_includes_ollama_without_api_key() -> None:
+    providers = [
+        SimpleNamespace(
+            id=1,
+            name="Ollama (Local)",
+            kind="openai",
+            enabled=True,
+            api_key="",
+            base_url="",
+            get_models=lambda: [{"id": "tiny", "name": "tiny", "enabled": True}],
+        )
+    ]
+    resolved = list_resolved_models(providers)
+    assert [(r["provider_name"], r["model"]) for r in resolved] == [("Ollama (Local)", "tiny")]
diff --git a/tests/unit/agents/test_provider_store.py b/tests/unit/agents/test_provider_store.py
index df4b3183..b14af7c6 100644
--- a/tests/unit/agents/test_provider_store.py
+++ b/tests/unit/agents/test_provider_store.py
@@ -322,6 +322,23 @@ def test_resolve_first_model_ref_returns_first_enabled_model(store: ProviderStor
     assert store.resolve_first_model_ref() == "p1/ready"
 
 
+def test_ollama_defaults_make_local_ref_usable(store: ProviderStore) -> None:
+    store._provider_repo.create(
+        name="Ollama (Local)",
+        kind="openai",
+        base_url="http://127.0.0.1:11434/v1",
+        api_key="",
+        models_json=json.dumps([{"id": "tiny", "name": "tiny", "enabled": True}]),
+    )
+    assert store.is_model_ref_usable("Ollama (Local)/tiny") is True
+    assert store.is_model_ref_usable("ollama/tiny") is True
+    configs = store.build_harness_configs()
+    assert configs[0].id == "Ollama (Local)"
+    assert configs[0].base_url == "http://127.0.0.1:11434/v1"
+    assert configs[0].api_key == "ollama"
+    assert [model.id for model in configs[0].models] == ["tiny"]
+
+
 def test_resolve_multimodal_model_ref_prefers_inferred_vision_model(store: ProviderStore) -> None:
     store._provider_repo.create(
         name="p",
diff --git a/tests/unit/i18n/test_stream.py b/tests/unit/i18n/test_stream.py
index 9e756757..53af13ad 100644
--- a/tests/unit/i18n/test_stream.py
+++ b/tests/unit/i18n/test_stream.py
@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 from octop.i18n.domains.stream import (
+    LOCAL_RUNTIME,
     MODEL_CALL_FAILED,
     RECURSION_LIMIT,
     STREAM_STALL,
@@ -90,6 +91,19 @@ def test_classify_recursion_limit() -> None:
     )
 
 
+def test_classify_local_runtime() -> None:
+    assert (
+        classify_stream_error_message("Failed to connect to ollama at http://127.0.0.1:11434")
+        == LOCAL_RUNTIME
+    )
+    assert classify_stream_error_message("Connection refused [Errno 111] 127.0.0.1:11434") == (
+        LOCAL_RUNTIME
+    )
+    text = format_stream_error("Failed to connect to ollama", "zh")
+    assert "Ollama" in text
+    assert "11434" in text
+
+
 def test_classify_model_call_failed_fallback() -> None:
     assert (
         classify_stream_error_message("Model call failed after 3 attempts with RuntimeError: boom")
diff --git a/tests/unit/test_local_register.py b/tests/unit/test_local_register.py
index 505efdb6..35ca7386 100644
--- a/tests/unit/test_local_register.py
+++ b/tests/unit/test_local_register.py
@@ -44,19 +44,34 @@ def _models() -> list[dict[str, object]]:
             return data if isinstance(data, list) else []
 
         row = SimpleNamespace(
-            id=1,
+            id=len(self.rows) + 1,
             name=kwargs["name"],
             kind=kwargs["kind"],
             base_url=kwargs.get("base_url"),
             api_key=kwargs.get("api_key"),
+            enabled=kwargs.get("enabled", True),
             models_json=models_json,
             get_models=_models,
         )
         self.rows.append(row)
-        return 1
+        return row.id
 
     def update(self, provider_id: int, **kwargs: object) -> None:
-        del provider_id, kwargs
+        row = next((item for item in self.rows if item.id == provider_id), None)
+        if row is None:
+            return
+        for key, value in kwargs.items():
+            setattr(row, key, value)
+        models_json = kwargs.get("models_json")
+        if isinstance(models_json, str):
+
+            def _models() -> list[dict[str, object]]:
+                import json
+
+                data = json.loads(models_json)
+                return data if isinstance(data, list) else []
+
+            row.get_models = _models
 
 
 def test_sanitize_model_name() -> None:
@@ -111,6 +126,7 @@ def _list() -> list[SimpleNamespace]:
     assert providers.rows[0].name == "Ollama (Local)"
     assert providers.rows[0].base_url == "http://127.0.0.1:11434/v1"
     assert "tiny.gguf" in settings.data.get("local_registered_weights", "")
+    assert settings.data.get("ollama_service_enabled") == "true"
 
 
 def test_register_ollama_source_skips_create(monkeypatch: pytest.MonkeyPatch) -> None:
@@ -223,6 +239,38 @@ def _boom() -> list[object]:
     assert created == [("tiny", "/tmp/tiny.gguf")]
 
 
+def test_register_enables_existing_local_service_row(monkeypatch: pytest.MonkeyPatch) -> None:
+    import json
+
+    monkeypatch.setattr(
+        "octop.infra.agents.providers.local_register.OllamaModelManager.list_models",
+        lambda: [SimpleNamespace(name="llama3.2:1b", size=99)],
+    )
+    providers = _Providers()
+    providers.create(
+        name="Ollama (Local)",
+        kind="openai",
+        base_url="http://127.0.0.1:11434/v1",
+        api_key="ollama",
+        models_json=json.dumps([{"id": "preset", "name": "preset", "enabled": False}]),
+        enabled=False,
+    )
+    settings = _Settings()
+    result = register_local_weight(
+        path="",
+        name="llama3.2:1b",
+        source="ollama",
+        size=99,
+        provider_repo=providers,
+        settings_repo=settings,
+    )
+    assert result["ok"] is True
+    assert providers.rows[0].enabled is True
+    models = providers.rows[0].get_models()
+    assert any(item.get("id") == "llama3.2:1b" and item.get("enabled") for item in models)
+    assert settings.data.get("ollama_service_enabled") == "true"
+
+
 def test_remember_weight_keys_ollama_rows_by_name() -> None:
     settings = _Settings()
     remember_weight(settings, {"name": "a", "path": "", "source": "ollama"})