From e9f3e3a7dae79158f5d390a6713eb9569e9ee58e Mon Sep 17 00:00:00 2001
From: Steven Enamakel's Droid <enamakel.agent@tinyhumans.ai>
Date: Mon, 1 Jun 2026 20:36:29 -0700
Subject: [PATCH] fix(inference): normalize 0.0.0.0 to 127.0.0.1 in Ollama URL
 + add endpoint edit button (#3161)

---
 .../components/settings/panels/AIPanel.tsx    |  27 ++-
 .../panels/__tests__/AIPanel.test.tsx         | 101 +++++++++++
 app/src/lib/i18n/ar.ts                        |   2 +
 app/src/lib/i18n/bn.ts                        |   2 +
 app/src/lib/i18n/de.ts                        |   2 +
 app/src/lib/i18n/en.ts                        |   2 +
 app/src/lib/i18n/es.ts                        |   2 +
 app/src/lib/i18n/fr.ts                        |   2 +
 app/src/lib/i18n/hi.ts                        |   2 +
 app/src/lib/i18n/id.ts                        |   2 +
 app/src/lib/i18n/it.ts                        |   2 +
 app/src/lib/i18n/ko.ts                        |   2 +
 app/src/lib/i18n/pl.ts                        |   2 +
 app/src/lib/i18n/pt.ts                        |   2 +
 app/src/lib/i18n/ru.ts                        |   2 +
 app/src/lib/i18n/zh-CN.ts                     |   2 +
 .../api/__tests__/aiSettingsApi.test.ts       |  49 ++++++
 app/src/services/api/aiSettingsApi.ts         |   4 +-
 app/src/utils/ollamaUrlValidation.test.ts     |   6 +
 app/src/utils/ollamaUrlValidation.ts          |  11 +-
 src/openhuman/about_app/catalog_data.rs       |   2 +-
 src/openhuman/config/ops.rs                   |  18 +-
 src/openhuman/config/ops_tests.rs             |  40 ++++-
 src/openhuman/config/schemas.rs               |  33 +++-
 src/openhuman/config/schemas_tests.rs         |  38 +++-
 src/openhuman/inference/local/mod.rs          |   4 +-
 src/openhuman/inference/local/ollama.rs       | 162 +++++++++++++++++-
 .../inference/local/service/assets.rs         |  24 +--
 .../inference/local/service/bootstrap.rs      |   2 +-
 .../inference/local/service/ollama_admin.rs   |  30 +++-
 .../inference/local/service/vision_embed.rs   |   4 +-
 src/openhuman/inference/schemas.rs            |  25 ++-
 src/openhuman/inference/schemas_tests.rs      |  14 ++
 tests/json_rpc_e2e.rs                         |  60 +++++++
 tests/memory_sync_sources_raw_coverage_e2e.rs |   3 +
 35 files changed, 634 insertions(+), 51 deletions(-)

diff --git a/app/src/components/settings/panels/AIPanel.tsx b/app/src/components/settings/panels/AIPanel.tsx
index 43a244e88..8afe1bf3a 100644
--- a/app/src/components/settings/panels/AIPanel.tsx
+++ b/app/src/components/settings/panels/AIPanel.tsx
@@ -9,7 +9,7 @@
  * per row, so the resolved provider+model is always rendered inline.
  */
 import { useCallback, useEffect, useMemo, useRef, useState } from 'react';
-import { LuCheck, LuCircleAlert, LuKeyRound } from 'react-icons/lu';
+import { LuCheck, LuCircleAlert, LuKeyRound, LuPencil } from 'react-icons/lu';
 
 import { listConnections as listComposioConnections } from '../../../lib/composio/composioApi';
 import type { ComposioConnection } from '../../../lib/composio/types';
@@ -582,6 +582,7 @@ const ProviderKeyDialog = ({
   slug,
   label,
   isLocalRuntime,
+  initialValue,
   oauthAction,
   onCancel,
   onSubmit,
@@ -590,6 +591,8 @@ const ProviderKeyDialog = ({
   label: string;
   /** When true, render an "Endpoint URL" field instead of API key. */
   isLocalRuntime: boolean;
+  /** Pre-populate the field when editing an existing provider's endpoint. */
+  initialValue?: string;
   oauthAction?: { label: string; description?: string; onClick: () => Promise<void> | void } | null;
   onCancel: () => void;
   /** Returns the entered value. For local runtimes this is the endpoint URL;
@@ -597,7 +600,9 @@ const ProviderKeyDialog = ({
   onSubmit: (value: string) => Promise<void> | void;
 }) => {
   const { t } = useT();
-  const [value, setValue] = useState<string>(isLocalRuntime ? defaultEndpointFor(slug) : '');
+  const [value, setValue] = useState<string>(
+    initialValue ?? (isLocalRuntime ? defaultEndpointFor(slug) : '')
+  );
   const [phase, setPhase] = useState<'idle' | 'saving' | 'oauth'>('idle');
   const [error, setError] = useState<string | null>(null);
   const busy = phase !== 'idle';
@@ -2884,6 +2889,19 @@ const AIPanel = ({ embedded = false }: AIPanelProps = {}) => {
                     key={localKind}
                     className={`inline-flex items-center gap-2 rounded-full px-2.5 py-1 text-xs font-medium ring-1 transition-colors ${tone}`}>
                     <span>{label}</span>
+                    {enabled && (
+                      <button
+                        type="button"
+                        aria-label={t('settings.ai.editEndpoint')}
+                        title={t('settings.ai.editEndpoint')}
+                        onClick={() => {
+                          setKeyDialogFor(localKind);
+                          setPendingLocalLabel(label);
+                        }}
+                        className="rounded p-0.5 hover:bg-black/10 dark:hover:bg-white/10 transition-colors">
+                        <LuPencil className="h-3 w-3" />
+                      </button>
+                    )}
                     <button
                       type="button"
                       role="switch"
@@ -3245,6 +3263,11 @@ const AIPanel = ({ embedded = false }: AIPanelProps = {}) => {
           slug={keyDialogFor}
           label={pendingLocalLabel ?? BUILTIN_PROVIDER_META[keyDialogFor]?.label ?? keyDialogFor}
           isLocalRuntime={Boolean(pendingLocalLabel)}
+          initialValue={
+            pendingLocalLabel
+              ? (draft.cloudProviders.find(cp => cp.slug === keyDialogFor)?.endpoint ?? undefined)
+              : undefined
+          }
           oauthAction={
             keyDialogFor === 'openrouter' && !pendingLocalLabel
               ? {
diff --git a/app/src/components/settings/panels/__tests__/AIPanel.test.tsx b/app/src/components/settings/panels/__tests__/AIPanel.test.tsx
index 72cef58e5..3fc4a4222 100644
--- a/app/src/components/settings/panels/__tests__/AIPanel.test.tsx
+++ b/app/src/components/settings/panels/__tests__/AIPanel.test.tsx
@@ -900,6 +900,48 @@ describe('AIPanel', () => {
     });
   });
 
+  it('passes Ollama 0.0.0.0 endpoint through to the Rust normalizer', async () => {
+    vi.mocked(loadAISettings).mockResolvedValue({ ...baseSettings, cloudProviders: [] });
+    renderWithProviders(<AIPanel />);
+    await waitFor(() =>
+      expect(screen.getByRole('switch', { name: /Connect Ollama/i })).toBeInTheDocument()
+    );
+    fireEvent.click(screen.getByRole('switch', { name: /Connect Ollama/i }));
+    const dialog = await screen.findByRole('dialog', { name: /Connect Ollama/i });
+
+    fireEvent.change(within(dialog).getByLabelText(/Endpoint URL/i), {
+      target: { value: 'http://0.0.0.0:11434' },
+    });
+    fireEvent.click(within(dialog).getByRole('button', { name: /^Save$/i }));
+
+    await waitFor(() => expect(openhumanUpdateLocalAiSettingsMock).toHaveBeenCalled());
+    const [arg] = vi.mocked(openhumanUpdateLocalAiSettingsMock).mock.calls[0];
+    expect(arg).toMatchObject({ base_url: 'http://0.0.0.0:11434' });
+  });
+
+  it('lets users edit an existing Ollama endpoint from the provider chip', async () => {
+    vi.mocked(loadAISettings).mockResolvedValue({
+      ...baseSettings,
+      cloudProviders: [
+        {
+          id: 'p_ollama_1',
+          slug: 'ollama',
+          label: 'Ollama',
+          endpoint: 'http://127.0.0.1:11434/v1',
+          auth_style: 'none' as const,
+          has_api_key: true,
+        },
+      ],
+    });
+    renderWithProviders(<AIPanel />);
+    const editButton = await screen.findByRole('button', { name: /Edit endpoint/i });
+    fireEvent.click(editButton);
+
+    const dialog = await screen.findByRole('dialog', { name: /Connect Ollama/i });
+    const urlInput = within(dialog).getByLabelText(/Endpoint URL/i) as HTMLInputElement;
+    expect(urlInput.value).toBe('http://127.0.0.1:11434/v1');
+  });
+
   it('LM Studio save persists the local_ai provider and endpoint', async () => {
     vi.mocked(loadAISettings).mockResolvedValue({ ...baseSettings, cloudProviders: [] });
     renderWithProviders(<AIPanel />);
@@ -924,6 +966,65 @@ describe('AIPanel', () => {
     });
   });
 
+  // ─── local runtime: edit endpoint button on enabled chip ────────────────────
+
+  it('shows an edit-endpoint button on enabled Ollama chip', async () => {
+    const settingsWithOllama = {
+      ...baseSettings,
+      cloudProviders: [
+        ...baseSettings.cloudProviders,
+        {
+          id: 'p_ollama_1',
+          slug: 'ollama',
+          label: 'Ollama',
+          endpoint: 'http://192.168.1.5:11434/v1',
+          auth_style: 'bearer' as const,
+          has_api_key: false,
+        },
+      ],
+    };
+    vi.mocked(loadAISettings).mockResolvedValue(settingsWithOllama);
+    renderWithProviders(<AIPanel />);
+
+    const editBtn = await screen.findByRole('button', { name: /Edit endpoint/i });
+    expect(editBtn).toBeInTheDocument();
+  });
+
+  it('edit-endpoint button opens the dialog pre-populated with the saved URL', async () => {
+    const settingsWithOllama = {
+      ...baseSettings,
+      cloudProviders: [
+        ...baseSettings.cloudProviders,
+        {
+          id: 'p_ollama_1',
+          slug: 'ollama',
+          label: 'Ollama',
+          endpoint: 'http://192.168.1.5:11434/v1',
+          auth_style: 'bearer' as const,
+          has_api_key: false,
+        },
+      ],
+    };
+    vi.mocked(loadAISettings).mockResolvedValue(settingsWithOllama);
+    renderWithProviders(<AIPanel />);
+
+    fireEvent.click(await screen.findByRole('button', { name: /Edit endpoint/i }));
+
+    const dialog = await screen.findByRole('dialog', { name: /Connect Ollama/i });
+    const urlInput = within(dialog).getByLabelText(/Endpoint URL/i) as HTMLInputElement;
+    expect(urlInput.value).toBe('http://192.168.1.5:11434/v1');
+  });
+
+  it('does not show an edit-endpoint button when Ollama is disabled', async () => {
+    vi.mocked(loadAISettings).mockResolvedValue({ ...baseSettings, cloudProviders: [] });
+    renderWithProviders(<AIPanel />);
+
+    await waitFor(() =>
+      expect(screen.getByRole('switch', { name: /Connect Ollama/i })).toBeInTheDocument()
+    );
+    expect(screen.queryByRole('button', { name: /Edit endpoint/i })).not.toBeInTheDocument();
+  });
+
   // ─── Custom routing dialog: per-workload temperature override ───────────────
 
   it('Custom routing dialog saves the routing change immediately from the modal', async () => {
diff --git a/app/src/lib/i18n/ar.ts b/app/src/lib/i18n/ar.ts
index 16906552f..32bdad73e 100644
--- a/app/src/lib/i18n/ar.ts
+++ b/app/src/lib/i18n/ar.ts
@@ -2752,7 +2752,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'مجموعة عبء عمل المحادثة',
   'settings.ai.disconnectProvider': 'قطع الاتصال {label}',
   'settings.ai.connectProviderLabel': 'الاتصال {label}',
+  'settings.ai.editProviderEndpoint': 'تعديل نقطة النهاية {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'تعديل نقطة الاتصال',
   'settings.ai.endpointUrlLabel': 'نقطة النهاية URL',
   'settings.ai.localRuntimeHelper':
     'حيث يمكن الوصول إلى {label}. الإعداد الافتراضي هو المضيف المحلي؛ وجّه هذا إلى مضيف بعيد (مثلًا http://10.0.0.4:11434/v1) لاستخدام نسخة مشتركة.',
diff --git a/app/src/lib/i18n/bn.ts b/app/src/lib/i18n/bn.ts
index c088aa027..380616ccb 100644
--- a/app/src/lib/i18n/bn.ts
+++ b/app/src/lib/i18n/bn.ts
@@ -2803,7 +2803,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'ওয়ার্কলোড গ্রুপ চ্যাট',
   'settings.ai.disconnectProvider': 'সংযোগ বিচ্ছিন্ন করুন {label}',
   'settings.ai.connectProviderLabel': 'সংযোগ {label}',
+  'settings.ai.editProviderEndpoint': '{label} শেষবিন্দু সম্পাদনা করুন',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'শেষবিন্দু সম্পাদনা করুন',
   'settings.ai.endpointUrlLabel': 'শেষবিন্দু URL',
   'settings.ai.localRuntimeHelper':
     'xqxqx সহযোগে সংযোগ ব্যবস্থা করা হবে। ডিফল্ট স্থানীয় হোস্ট; দূরবর্তী হোস্টটিকে এই অবস্থায় দেখা যাবে (যেমন, xq1xqxqx ব্যবহার করা হচ্ছে)।',
diff --git a/app/src/lib/i18n/de.ts b/app/src/lib/i18n/de.ts
index 1de6591a3..c515e38cb 100644
--- a/app/src/lib/i18n/de.ts
+++ b/app/src/lib/i18n/de.ts
@@ -2873,7 +2873,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Workload-Gruppenchat',
   'settings.ai.disconnectProvider': 'Trennen {label}',
   'settings.ai.connectProviderLabel': 'Verbinden {label}',
+  'settings.ai.editProviderEndpoint': 'Endpunkt für {label} bearbeiten',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Endpunkt bearbeiten',
   'settings.ai.endpointUrlLabel': 'Endpunkt URL',
   'settings.ai.localRuntimeHelper':
     'Wo {label} erreichbar ist. Die Standardeinstellung ist localhost. Richten Sie dies auf einen Remote-Host (z. B. http://10.0.0.4:11434/v1), um eine freigegebene Instanz zu verwenden.',
diff --git a/app/src/lib/i18n/en.ts b/app/src/lib/i18n/en.ts
index 67bbbbba4..ab85b6715 100644
--- a/app/src/lib/i18n/en.ts
+++ b/app/src/lib/i18n/en.ts
@@ -3067,7 +3067,9 @@ const en: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Chat and conversations',
   'settings.ai.disconnectProvider': 'Disconnect {label}',
   'settings.ai.connectProviderLabel': 'Connect {label}',
+  'settings.ai.editProviderEndpoint': 'Edit {label} endpoint',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Edit endpoint',
   'settings.ai.endpointUrlLabel': 'Endpoint URL',
   'settings.ai.localRuntimeHelper':
     'Where {label} is reachable. Default is localhost; point this at a remote host (for example, http://10.0.0.4:11434/v1) to use a shared instance.',
diff --git a/app/src/lib/i18n/es.ts b/app/src/lib/i18n/es.ts
index 4a8916e86..fbf428496 100644
--- a/app/src/lib/i18n/es.ts
+++ b/app/src/lib/i18n/es.ts
@@ -2855,7 +2855,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Grupo de carga de trabajo de chat',
   'settings.ai.disconnectProvider': 'Desconectar {label}',
   'settings.ai.connectProviderLabel': 'Conectar {label}',
+  'settings.ai.editProviderEndpoint': 'Editar punto de conexión de {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Editar punto de conexión',
   'settings.ai.endpointUrlLabel': 'Punto final URL',
   'settings.ai.localRuntimeHelper':
     'Donde {label} es accesible. Por defecto es localhost; apunte esto a un host remoto (por ejemplo, http://10.0.0.4:11434/v1) para usar una instancia compartida.',
diff --git a/app/src/lib/i18n/fr.ts b/app/src/lib/i18n/fr.ts
index 8a133f30e..24ff48ae7 100644
--- a/app/src/lib/i18n/fr.ts
+++ b/app/src/lib/i18n/fr.ts
@@ -2863,7 +2863,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Groupe de charge chat',
   'settings.ai.disconnectProvider': 'Déconnecter {label}',
   'settings.ai.connectProviderLabel': 'Connecter {label}',
+  'settings.ai.editProviderEndpoint': 'Modifier le point de terminaison {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Modifier le point de terminaison',
   'settings.ai.endpointUrlLabel': 'Point de terminaison URL',
   'settings.ai.localRuntimeHelper':
     "Là où {label} est accessible. Par défaut, c'est localhost; pointez ceci vers un hôte distant (par exemple, http://10.0.0.4:11434/v1) pour utiliser une instance partagée.",
diff --git a/app/src/lib/i18n/hi.ts b/app/src/lib/i18n/hi.ts
index 3ca93f11d..245983a51 100644
--- a/app/src/lib/i18n/hi.ts
+++ b/app/src/lib/i18n/hi.ts
@@ -2808,7 +2808,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'चैट वर्कलोड ग्रुप',
   'settings.ai.disconnectProvider': 'डिस्कनेक्ट करें {label}',
   'settings.ai.connectProviderLabel': 'कनेक्ट करें {label}',
+  'settings.ai.editProviderEndpoint': '{label} समापन बिंदु संपादित करें',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'समापन बिंदु संपादित करें',
   'settings.ai.endpointUrlLabel': 'समापन बिंदु URL',
   'settings.ai.localRuntimeHelper':
     'जहां {label} पहुंच योग्य है। डिफ़ॉल्ट स्थानीयहोस्ट है; इसे रिमोट होस्ट पर इंगित करें (उदाहरण के लिए, http://10.0.0.4:11434/v1) एक साझा उदाहरण का उपयोग करने के लिए)।',
diff --git a/app/src/lib/i18n/id.ts b/app/src/lib/i18n/id.ts
index 2333b1d27..225e91304 100644
--- a/app/src/lib/i18n/id.ts
+++ b/app/src/lib/i18n/id.ts
@@ -2812,7 +2812,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Grup beban kerja chat',
   'settings.ai.disconnectProvider': 'Putuskan sambungan {label}',
   'settings.ai.connectProviderLabel': 'Sambungkan {label}',
+  'settings.ai.editProviderEndpoint': 'Ubah titik akhir {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Edit titik akhir',
   'settings.ai.endpointUrlLabel': 'Titik Akhir URL',
   'settings.ai.localRuntimeHelper':
     'Dimana {label} bisa dihubungi. Baku adalah localhost; arahkan ini ke host jarak jauh (misalnya, http://10.0.0.4:11434/v1) untuk memakai suatu contoh bersama.',
diff --git a/app/src/lib/i18n/it.ts b/app/src/lib/i18n/it.ts
index ec11ef652..b22df856b 100644
--- a/app/src/lib/i18n/it.ts
+++ b/app/src/lib/i18n/it.ts
@@ -2846,7 +2846,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Gruppo carico di lavoro chat',
   'settings.ai.disconnectProvider': 'Disconnetti {label}',
   'settings.ai.connectProviderLabel': 'Connetti {label}',
+  'settings.ai.editProviderEndpoint': 'Modifica punto finale di {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': "Modifica l'endpoint",
   'settings.ai.endpointUrlLabel': "URL dell'endpoint",
   'settings.ai.localRuntimeHelper':
     "Dove {label} è raggiungibile. Il valore predefinito è localhost; punta questo a un host remoto (per esempio, http://10.0.0.4:11434/v1) per utilizzare un'istanza condivisa.",
diff --git a/app/src/lib/i18n/ko.ts b/app/src/lib/i18n/ko.ts
index 0dda26c3a..2a222045a 100644
--- a/app/src/lib/i18n/ko.ts
+++ b/app/src/lib/i18n/ko.ts
@@ -2786,7 +2786,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': '채팅 작업 그룹',
   'settings.ai.disconnectProvider': '{label} 연결 끊기',
   'settings.ai.connectProviderLabel': '{label} 연결',
+  'settings.ai.editProviderEndpoint': '{label} 엔드포인트 편집',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': '엔드포인트 편집',
   'settings.ai.endpointUrlLabel': '끝점 URL',
   'settings.ai.localRuntimeHelper':
     '{label}에 연결할 수 있는 위치입니다. 기본값은 localhost입니다. 공유 인스턴스를 사용하려면 원격 호스트(예: http://10.0.0.4:11434/v1)를 가리키세요.',
diff --git a/app/src/lib/i18n/pl.ts b/app/src/lib/i18n/pl.ts
index c6b095e70..3f58cb516 100644
--- a/app/src/lib/i18n/pl.ts
+++ b/app/src/lib/i18n/pl.ts
@@ -2852,7 +2852,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Grupa workloadów: czat',
   'settings.ai.disconnectProvider': 'Rozłącz {label}',
   'settings.ai.connectProviderLabel': 'Połącz {label}',
+  'settings.ai.editProviderEndpoint': 'Edytuj punkt końcowy {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Edytuj endpoint',
   'settings.ai.endpointUrlLabel': 'URL endpointu',
   'settings.ai.localRuntimeHelper':
     'Miejsce, w którym dostępny jest {label}. Domyślnie jest to localhost; wskaż zdalny host (na przykład http://10.0.0.4:11434/v1), aby użyć współdzielonej instancji.',
diff --git a/app/src/lib/i18n/pt.ts b/app/src/lib/i18n/pt.ts
index 864602b1c..2037e20de 100644
--- a/app/src/lib/i18n/pt.ts
+++ b/app/src/lib/i18n/pt.ts
@@ -2851,7 +2851,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Grupo de carga de trabalho de chat',
   'settings.ai.disconnectProvider': 'Desconecte {label}',
   'settings.ai.connectProviderLabel': 'Conecte {label}',
+  'settings.ai.editProviderEndpoint': 'Editar ponto de extremidade de {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Editar ponto de extremidade',
   'settings.ai.endpointUrlLabel': 'Ponto de extremidade URL',
   'settings.ai.localRuntimeHelper':
     'Onde {label} é acessível. O padrão é localhost; aponte isso para um host remoto (por exemplo, http://10.0.0.4:11434/v1) para usar uma instância compartilhada.',
diff --git a/app/src/lib/i18n/ru.ts b/app/src/lib/i18n/ru.ts
index 9beafc55b..18301d55c 100644
--- a/app/src/lib/i18n/ru.ts
+++ b/app/src/lib/i18n/ru.ts
@@ -2822,7 +2822,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': 'Чат',
   'settings.ai.disconnectProvider': 'Отключить {label}',
   'settings.ai.connectProviderLabel': 'Подключить {label}',
+  'settings.ai.editProviderEndpoint': 'Изменить конечную точку {label}',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': 'Изменить конечную точку',
   'settings.ai.endpointUrlLabel': 'Конечная точка URL',
   'settings.ai.localRuntimeHelper':
     'Где доступен {label}. По умолчанию используется локальный хост; укажите это на удаленном хосте (например, http://10.0.0.4:11434/v1), чтобы использовать общий экземпляр.',
diff --git a/app/src/lib/i18n/zh-CN.ts b/app/src/lib/i18n/zh-CN.ts
index 2c2e43c2a..1094d9919 100644
--- a/app/src/lib/i18n/zh-CN.ts
+++ b/app/src/lib/i18n/zh-CN.ts
@@ -2674,7 +2674,9 @@ const messages: TranslationMap = {
   'settings.ai.workloadGroupChat': '对话工作负载',
   'settings.ai.disconnectProvider': '断开 {label}',
   'settings.ai.connectProviderLabel': '连接 {label}',
+  'settings.ai.editProviderEndpoint': '编辑 {label} 端点',
   'settings.ai.defaultLocalEndpoint': 'http://localhost:11434/v1',
+  'settings.ai.editEndpoint': '编辑端点',
   'settings.ai.endpointUrlLabel': '端点 URL',
   'settings.ai.localRuntimeHelper':
     '{label} 可访问的位置。默认是 localhost；如需使用共享实例，可指向远程主机（例如 http://10.0.0.4:11434/v1）。',
diff --git a/app/src/services/api/__tests__/aiSettingsApi.test.ts b/app/src/services/api/__tests__/aiSettingsApi.test.ts
index e4ee48ef9..9dbc8413c 100644
--- a/app/src/services/api/__tests__/aiSettingsApi.test.ts
+++ b/app/src/services/api/__tests__/aiSettingsApi.test.ts
@@ -411,6 +411,29 @@ describe('loadAISettings', () => {
     expect(settings.cloudProviders[0].has_api_key).toBe(false);
   });
 
+  it('keeps local runtime endpoint providers so the AI panel can edit them', async () => {
+    mockOpenhumanGetClientConfig.mockResolvedValue(
+      makeClientConfigResult({
+        cloud_providers: [
+          {
+            id: 'p_ollama_1',
+            slug: 'ollama',
+            label: 'Ollama',
+            endpoint: 'http://127.0.0.1:11434/v1',
+            auth_style: 'none',
+          },
+        ],
+      })
+    );
+    mockAuthListProviderCredentials.mockResolvedValue(makeAuthProfileResult([]));
+
+    const settings = await loadAISettings();
+
+    expect(settings.cloudProviders).toHaveLength(1);
+    expect(settings.cloudProviders[0].slug).toBe('ollama');
+    expect(settings.cloudProviders[0].endpoint).toBe('http://127.0.0.1:11434/v1');
+  });
+
   it('includes two cloud providers with correct labels and endpoints', async () => {
     mockOpenhumanGetClientConfig.mockResolvedValue(
       makeClientConfigResult({
@@ -602,6 +625,32 @@ describe('saveAISettings', () => {
     expect(patch.cloud_providers![0]).not.toHaveProperty('has_api_key');
   });
 
+  it('preserves local runtime providers in the cloud_providers payload', async () => {
+    const prev = makeSettings({ cloudProviders: [] });
+    const next = makeSettings({
+      cloudProviders: [
+        {
+          id: 'p_ollama_1',
+          slug: 'ollama',
+          label: 'Ollama',
+          endpoint: 'http://127.0.0.1:11434/v1',
+          auth_style: 'none',
+          has_api_key: true,
+        },
+      ],
+    });
+
+    await saveAISettings(prev, next);
+
+    const patch = mockOpenhumanUpdateModelSettings.mock.calls[0][0];
+    expect(patch.cloud_providers).toHaveLength(1);
+    expect(patch.cloud_providers![0]).toMatchObject({
+      slug: 'ollama',
+      endpoint: 'http://127.0.0.1:11434/v1',
+      auth_style: 'none',
+    });
+  });
+
   it('preserves auth_style through save round-trip for anthropic', async () => {
     const anthropicProvider = {
       id: 'p_anthropic_1',
diff --git a/app/src/services/api/aiSettingsApi.ts b/app/src/services/api/aiSettingsApi.ts
index 9401dadce..0b55af952 100644
--- a/app/src/services/api/aiSettingsApi.ts
+++ b/app/src/services/api/aiSettingsApi.ts
@@ -215,7 +215,7 @@ export async function loadAISettings(): Promise<AISettings> {
   );
 
   const cloudProviders: CloudProviderView[] = config.cloud_providers
-    .filter(p => !['', 'cloud', 'openhuman', 'ollama', 'pid'].includes(p.slug.trim()))
+    .filter(p => !['', 'cloud', 'openhuman', 'pid'].includes(p.slug.trim()))
     .map(p => {
       const newKey = authKeyForSlug(p.slug).toLowerCase();
       const legacyKey = p.slug.toLowerCase();
@@ -282,7 +282,7 @@ export async function saveAISettings(prev: AISettings, next: AISettings): Promis
     })
   ) {
     patch.cloud_providers = next.cloudProviders
-      .filter(p => !['', 'cloud', 'openhuman', 'ollama', 'pid'].includes(p.slug.trim()))
+      .filter(p => !['', 'cloud', 'openhuman', 'pid'].includes(p.slug.trim()))
       .map(({ id, slug, label, endpoint, auth_style }) => ({
         id,
         slug,
diff --git a/app/src/utils/ollamaUrlValidation.test.ts b/app/src/utils/ollamaUrlValidation.test.ts
index adaae276a..2f799bc96 100644
--- a/app/src/utils/ollamaUrlValidation.test.ts
+++ b/app/src/utils/ollamaUrlValidation.test.ts
@@ -71,6 +71,12 @@ describe('validateOllamaUrl', () => {
     expect(result.normalized).toBe('https://example.com');
   });
 
+  it('normalizes server bind addresses to client loopback addresses', () => {
+    const result = validateOllamaUrl('http://0.0.0.0:11434');
+    expect(result.valid).toBe(true);
+    expect(result.normalized).toBe('http://localhost:11434');
+  });
+
   it('rejects a URL that starts with http:// but is not parseable', () => {
     const result = validateOllamaUrl('http:// has spaces');
     expect(result.valid).toBe(false);
diff --git a/app/src/utils/ollamaUrlValidation.ts b/app/src/utils/ollamaUrlValidation.ts
index 50bccb70d..18a90466f 100644
--- a/app/src/utils/ollamaUrlValidation.ts
+++ b/app/src/utils/ollamaUrlValidation.ts
@@ -41,8 +41,15 @@ export function validateOllamaUrl(raw: string): OllamaUrlValidationResult {
   if (parsed.hash) {
     return { valid: false, error: 'URL must not contain a fragment' };
   }
-  // Normalize to scheme://host[:port]
+  // Normalize to scheme://host[:port]; rewrite bind-all addresses to loopback.
+  // JS URL API strips brackets from IPv6 hostnames: new URL('http://[::]:11434').hostname === '::'
+  const hostname =
+    parsed.hostname === '0.0.0.0'
+      ? 'localhost'
+      : parsed.hostname === '::'
+        ? '[::1]'
+        : parsed.hostname;
   const port = parsed.port ? `:${parsed.port}` : '';
-  const normalized = `${parsed.protocol}//${parsed.hostname}${port}`;
+  const normalized = `${parsed.protocol}//${hostname}${port}`;
   return { valid: true, normalized };
 }
diff --git a/src/openhuman/about_app/catalog_data.rs b/src/openhuman/about_app/catalog_data.rs
index 8389d6fca..af6c126bf 100644
--- a/src/openhuman/about_app/catalog_data.rs
+++ b/src/openhuman/about_app/catalog_data.rs
@@ -668,7 +668,7 @@ pub(super) const CAPABILITIES: &[Capability] = &[
         domain: "local_ai",
         category: CapabilityCategory::LocalAI,
         description: "Select Ollama or LM Studio as the local model provider and configure the local server endpoint.",
-        how_to: "Settings > Local AI Model",
+        how_to: "Settings > AI > providers, or Settings > Local AI Model > Ollama server URL",
         status: CapabilityStatus::Beta,
         privacy: None,
     },
diff --git a/src/openhuman/config/ops.rs b/src/openhuman/config/ops.rs
index 7f7c585e3..9e078ba99 100644
--- a/src/openhuman/config/ops.rs
+++ b/src/openhuman/config/ops.rs
@@ -458,7 +458,7 @@ pub struct LocalAiSettingsPatch {
     /// behaviour without needing to apply a preset first.
     pub opt_in_confirmed: Option<bool>,
     pub provider: Option<String>,
-    pub base_url: Option<String>,
+    pub base_url: Option<Option<String>>,
     pub model_id: Option<String>,
     pub chat_model_id: Option<String>,
     pub usage_embeddings: Option<bool>,
@@ -1316,10 +1316,18 @@ pub async fn apply_local_ai_settings(
             crate::openhuman::inference::local::provider::normalize_provider(&provider);
     }
     if let Some(base_url) = update.base_url {
-        config.local_ai.base_url = if base_url.trim().is_empty() {
-            None
-        } else {
-            Some(base_url.trim().to_string())
+        config.local_ai.base_url = match base_url {
+            None => None,
+            Some(base_url) if base_url.trim().is_empty() => None,
+            Some(base_url)
+                if crate::openhuman::inference::local::provider::provider_from_config(config)
+                    == crate::openhuman::inference::local::provider::LocalAiProvider::Ollama =>
+            {
+                Some(crate::openhuman::inference::local::validate_ollama_url(
+                    &base_url,
+                )?)
+            }
+            Some(base_url) => Some(base_url.trim().trim_end_matches('/').to_string()),
         };
     }
     if let Some(model_id) = update.model_id {
diff --git a/src/openhuman/config/ops_tests.rs b/src/openhuman/config/ops_tests.rs
index f5946f7e9..211cf1736 100644
--- a/src/openhuman/config/ops_tests.rs
+++ b/src/openhuman/config/ops_tests.rs
@@ -802,7 +802,7 @@ async fn apply_local_ai_settings_updates_lm_studio_provider_fields() {
         runtime_enabled: Some(true),
         opt_in_confirmed: Some(true),
         provider: Some("lm-studio".into()),
-        base_url: Some(" http://localhost:1234/v1/ ".into()),
+        base_url: Some(Some(" http://localhost:1234/v1/ ".into())),
         model_id: Some(" local-default ".into()),
         chat_model_id: Some(" local-chat ".into()),
         usage_embeddings: Some(true),
@@ -820,7 +820,7 @@ async fn apply_local_ai_settings_updates_lm_studio_provider_fields() {
     assert_eq!(cfg.local_ai.provider, "lm_studio");
     assert_eq!(
         cfg.local_ai.base_url.as_deref(),
-        Some("http://localhost:1234/v1/")
+        Some("http://localhost:1234/v1")
     );
     assert_eq!(cfg.local_ai.model_id, "local-default");
     assert_eq!(cfg.local_ai.chat_model_id, "local-chat");
@@ -832,7 +832,7 @@ async fn apply_local_ai_settings_updates_lm_studio_provider_fields() {
 
     let clear_and_fallback = LocalAiSettingsPatch {
         provider: Some("unknown-provider".into()),
-        base_url: Some("   ".into()),
+        base_url: Some(Some("   ".into())),
         model_id: Some("   ".into()),
         chat_model_id: Some("".into()),
         ..LocalAiSettingsPatch::default()
@@ -847,6 +847,40 @@ async fn apply_local_ai_settings_updates_lm_studio_provider_fields() {
     assert_eq!(cfg.local_ai.chat_model_id, "");
 }
 
+#[tokio::test]
+async fn apply_local_ai_settings_normalizes_ollama_unspecified_host_and_allows_null_clear() {
+    let tmp = tempdir().unwrap();
+    let mut cfg = tmp_config(&tmp);
+
+    apply_local_ai_settings(
+        &mut cfg,
+        LocalAiSettingsPatch {
+            provider: Some("ollama".into()),
+            base_url: Some(Some("http://0.0.0.0:11434/api/tags".into())),
+            ..LocalAiSettingsPatch::default()
+        },
+    )
+    .await
+    .expect("apply ollama base url");
+
+    assert_eq!(
+        cfg.local_ai.base_url.as_deref(),
+        Some("http://localhost:11434")
+    );
+
+    apply_local_ai_settings(
+        &mut cfg,
+        LocalAiSettingsPatch {
+            base_url: Some(None),
+            ..LocalAiSettingsPatch::default()
+        },
+    )
+    .await
+    .expect("clear ollama base url");
+
+    assert!(cfg.local_ai.base_url.is_none());
+}
+
 #[tokio::test]
 async fn apply_analytics_settings_updates_enabled() {
     let tmp = tempdir().unwrap();
diff --git a/src/openhuman/config/schemas.rs b/src/openhuman/config/schemas.rs
index 441f21727..a127b5e9c 100644
--- a/src/openhuman/config/schemas.rs
+++ b/src/openhuman/config/schemas.rs
@@ -1,4 +1,4 @@
-use serde::de::DeserializeOwned;
+use serde::de::{DeserializeOwned, Deserializer};
 use serde::Deserialize;
 use serde_json::{Map, Value};
 
@@ -142,7 +142,8 @@ struct LocalAiSettingsUpdate {
     /// having to also apply a tier preset.
     opt_in_confirmed: Option<bool>,
     provider: Option<String>,
-    base_url: Option<String>,
+    #[serde(default, deserialize_with = "deserialize_present_json")]
+    base_url: Option<Value>,
     model_id: Option<String>,
     chat_model_id: Option<String>,
     usage_embeddings: Option<bool>,
@@ -685,9 +686,9 @@ pub fn schemas(function: &str) -> ControllerSchema {
                     "provider",
                     "Local provider identifier. Supported values: ollama, lm_studio.",
                 ),
-                optional_string(
+                optional_json(
                     "base_url",
-                    "Provider base URL. For LM Studio this defaults to http://localhost:1234/v1.",
+                    "Provider base URL string, or null to clear. For LM Studio this defaults to http://localhost:1234/v1.",
                 ),
                 optional_string("model_id", "Default local chat model identifier."),
                 optional_string("chat_model_id", "Local chat model identifier."),
@@ -1345,11 +1346,17 @@ fn handle_update_browser_settings(params: Map<String, Value>) -> ControllerFutur
 fn handle_update_local_ai_settings(params: Map<String, Value>) -> ControllerFuture {
     Box::pin(async move {
         let update = deserialize_params::<LocalAiSettingsUpdate>(params)?;
+        let base_url = match update.base_url {
+            None => None,
+            Some(Value::Null) => Some(None),
+            Some(Value::String(value)) => Some(Some(value)),
+            Some(_) => return Err("invalid params: base_url must be a string or null".to_string()),
+        };
         let patch = config_rpc::LocalAiSettingsPatch {
             runtime_enabled: update.runtime_enabled,
             opt_in_confirmed: update.opt_in_confirmed,
             provider: update.provider,
-            base_url: update.base_url,
+            base_url,
             model_id: update.model_id,
             chat_model_id: update.chat_model_id,
             usage_embeddings: update.usage_embeddings,
@@ -1654,6 +1661,13 @@ fn deserialize_params<T: DeserializeOwned>(params: Map<String, Value>) -> Result
     serde_json::from_value(Value::Object(params)).map_err(|e| format!("invalid params: {e}"))
 }
 
+fn deserialize_present_json<'de, D>(deserializer: D) -> Result<Option<Value>, D::Error>
+where
+    D: Deserializer<'de>,
+{
+    Value::deserialize(deserializer).map(Some)
+}
+
 fn optional_string(name: &'static str, comment: &'static str) -> FieldSchema {
     FieldSchema {
         name,
@@ -1663,6 +1677,15 @@ fn optional_string(name: &'static str, comment: &'static str) -> FieldSchema {
     }
 }
 
+fn optional_json(name: &'static str, comment: &'static str) -> FieldSchema {
+    FieldSchema {
+        name,
+        ty: TypeSchema::Option(Box::new(TypeSchema::Json)),
+        comment,
+        required: false,
+    }
+}
+
 fn required_string(name: &'static str, comment: &'static str) -> FieldSchema {
     FieldSchema {
         name,
diff --git a/src/openhuman/config/schemas_tests.rs b/src/openhuman/config/schemas_tests.rs
index 4eacf6e07..a2aa6e80b 100644
--- a/src/openhuman/config/schemas_tests.rs
+++ b/src/openhuman/config/schemas_tests.rs
@@ -106,6 +106,16 @@ fn optional_string_builds_option_string_field() {
     }
 }
 
+#[test]
+fn optional_json_builds_option_json_field() {
+    let f = optional_json("payload", "json payload");
+    assert!(!f.required);
+    match &f.ty {
+        TypeSchema::Option(inner) => assert!(matches!(**inner, TypeSchema::Json)),
+        other => panic!("expected Option<Json>, got {other:?}"),
+    }
+}
+
 #[test]
 fn optional_bool_builds_option_bool_field() {
     let f = optional_bool("enabled", "Whether enabled");
@@ -206,13 +216,39 @@ fn deserialize_params_parses_local_ai_settings_update() {
     assert_eq!(out.runtime_enabled, Some(true));
     assert_eq!(out.opt_in_confirmed, Some(true));
     assert_eq!(out.provider.as_deref(), Some("lm_studio"));
-    assert_eq!(out.base_url.as_deref(), Some("http://localhost:1234/v1"));
+    assert_eq!(
+        out.base_url.as_ref().and_then(Value::as_str),
+        Some("http://localhost:1234/v1")
+    );
     assert_eq!(out.model_id.as_deref(), Some("local-default"));
     assert_eq!(out.chat_model_id.as_deref(), Some("local-chat"));
     assert_eq!(out.usage_embeddings, Some(true));
     assert_eq!(out.usage_subconscious, Some(false));
 }
 
+#[test]
+fn deserialize_params_preserves_local_ai_base_url_null() {
+    let mut m = Map::new();
+    m.insert("base_url".into(), Value::Null);
+
+    let out: LocalAiSettingsUpdate = deserialize_params(m).unwrap();
+    assert!(out.base_url.as_ref().is_some_and(Value::is_null));
+}
+
+#[test]
+fn update_local_ai_settings_schema_allows_json_base_url() {
+    let schema = schemas("update_local_ai_settings");
+    let field = schema
+        .inputs
+        .iter()
+        .find(|field| field.name == "base_url")
+        .expect("base_url field");
+    match &field.ty {
+        TypeSchema::Option(inner) => assert!(matches!(**inner, TypeSchema::Json)),
+        other => panic!("expected Option<Json>, got {other:?}"),
+    }
+}
+
 #[test]
 fn deserialize_params_parses_workspace_onboarding_flag_params() {
     let out: WorkspaceOnboardingFlagParams = deserialize_params(Map::new()).unwrap();
diff --git a/src/openhuman/inference/local/mod.rs b/src/openhuman/inference/local/mod.rs
index eca1ba14e..d4913e0ad 100644
--- a/src/openhuman/inference/local/mod.rs
+++ b/src/openhuman/inference/local/mod.rs
@@ -38,7 +38,9 @@ mod ollama;
 mod process_util;
 pub(crate) mod provider;
 pub(crate) use model_requirements::{evaluate_context, ContextEligibility, MIN_CONTEXT_TOKENS};
-pub(crate) use ollama::{ollama_base_url, ollama_base_url_from_config, OLLAMA_BASE_URL};
+pub(crate) use ollama::{
+    ollama_base_url, ollama_base_url_from_config, validate_ollama_url, OLLAMA_BASE_URL,
+};
 pub mod service;
 pub(crate) mod voice_install_common;
 
diff --git a/src/openhuman/inference/local/ollama.rs b/src/openhuman/inference/local/ollama.rs
index be297fb36..998bd2026 100644
--- a/src/openhuman/inference/local/ollama.rs
+++ b/src/openhuman/inference/local/ollama.rs
@@ -4,6 +4,34 @@ use serde::{Deserialize, Serialize};
 
 pub(crate) const DEFAULT_OLLAMA_BASE_URL: &str = "http://localhost:11434";
 
+/// Rewrite unspecified bind addresses (`0.0.0.0`, `[::]`) to their loopback
+/// equivalents (`127.0.0.1`, `[::1]`).  Ollama's default `OLLAMA_HOST` is
+/// `0.0.0.0:11434` — a valid *server-side* bind address but an invalid
+/// *client-side* connect target on Windows (and misleading on other OSes).
+fn normalize_unspecified_host(url: &str) -> String {
+    if let Ok(parsed) = reqwest::Url::parse(url) {
+        let replacement = match parsed.host() {
+            Some(url::Host::Ipv4(addr)) if addr.is_unspecified() => Some("localhost"),
+            Some(url::Host::Ipv6(addr)) if addr.is_unspecified() => Some("[::1]"),
+            _ => None,
+        };
+        if let Some(new_host) = replacement {
+            let scheme = parsed.scheme();
+            let port_suffix = parsed.port().map(|p| format!(":{p}")).unwrap_or_default();
+            let path = parsed.path().trim_end_matches('/');
+            let result = format!("{scheme}://{new_host}{port_suffix}{path}");
+            let result = result.trim_end_matches('/').to_string();
+            log::debug!(
+                "[local_ai] normalize_unspecified_host: rewrote {} -> {}",
+                redact_ollama_base_url(url),
+                redact_ollama_base_url(&result)
+            );
+            return result;
+        }
+    }
+    url.to_string()
+}
+
 /// Returns the effective Ollama base URL.
 ///
 /// Priority (highest to lowest):
@@ -11,11 +39,14 @@ pub(crate) const DEFAULT_OLLAMA_BASE_URL: &str = "http://localhost:11434";
 /// 2. `OLLAMA_HOST` — Ollama's own env var; normalized to a full URL by
 ///    prepending `http://` when no scheme is present.
 /// 3. [`DEFAULT_OLLAMA_BASE_URL`] — `http://localhost:11434`.
+///
+/// Unspecified bind addresses (`0.0.0.0`, `[::]`) are rewritten to their
+/// loopback equivalents so the URL is valid as a client connect target.
 pub(crate) fn ollama_base_url() -> String {
     if let Ok(url) = std::env::var("OPENHUMAN_OLLAMA_BASE_URL") {
         let trimmed = url.trim();
         if !trimmed.is_empty() {
-            return trimmed.trim_end_matches('/').to_string();
+            return normalize_unspecified_host(trimmed.trim_end_matches('/'));
         }
     }
 
@@ -28,7 +59,7 @@ pub(crate) fn ollama_base_url() -> String {
                 format!("http://{trimmed}")
             };
             log::debug!("[local_ai] ollama_base_url: using OLLAMA_HOST -> {url}");
-            return url;
+            return normalize_unspecified_host(&url);
         }
     }
 
@@ -43,15 +74,19 @@ pub(crate) fn ollama_base_url() -> String {
 /// 2. `OPENHUMAN_OLLAMA_BASE_URL` env var
 /// 3. `OLLAMA_HOST` env var
 /// 4. [`DEFAULT_OLLAMA_BASE_URL`]
+///
+/// Unspecified bind addresses (`0.0.0.0`, `[::]`) are rewritten to their
+/// loopback equivalents so the URL is valid as a client connect target.
 pub(crate) fn ollama_base_url_from_config(config: &crate::openhuman::config::Config) -> String {
     if let Some(ref url) = config.local_ai.base_url {
         let trimmed = url.trim().trim_end_matches('/');
         if !trimmed.is_empty() {
+            let normalized = normalize_unspecified_host(trimmed);
             log::debug!(
                 "[local_ai] ollama_base_url_from_config: using config base_url -> {}",
-                redact_ollama_base_url(trimmed)
+                redact_ollama_base_url(&normalized)
             );
-            return trimmed.to_string();
+            return normalized;
         }
     }
     let resolved = ollama_base_url();
@@ -101,6 +136,8 @@ pub(crate) fn validate_ollama_url(raw: &str) -> Result<String, String> {
     // Use the Host enum so IPv6 addresses are always re-bracketed correctly,
     // regardless of whether host_str() includes brackets in a given url-crate version.
     let host_formatted = match parsed.host() {
+        Some(url::Host::Ipv4(addr)) if addr.is_unspecified() => "localhost".to_string(),
+        Some(url::Host::Ipv6(addr)) if addr.is_unspecified() => "[::1]".to_string(),
         Some(url::Host::Ipv6(addr)) => format!("[{addr}]"),
         Some(h) => h.to_string(),
         None => String::new(),
@@ -683,6 +720,107 @@ mod tests {
         );
     }
 
+    // ── normalize_unspecified_host ──────────────────────────────────────
+
+    #[test]
+    fn normalize_rewrites_ipv4_unspecified() {
+        assert_eq!(
+            normalize_unspecified_host("http://0.0.0.0:11434"),
+            "http://localhost:11434"
+        );
+    }
+
+    #[test]
+    fn normalize_rewrites_ipv6_unspecified() {
+        assert_eq!(
+            normalize_unspecified_host("http://[::]:11434"),
+            "http://[::1]:11434"
+        );
+    }
+
+    #[test]
+    fn normalize_preserves_loopback() {
+        assert_eq!(
+            normalize_unspecified_host("http://127.0.0.1:11434"),
+            "http://127.0.0.1:11434"
+        );
+        assert_eq!(
+            normalize_unspecified_host("http://[::1]:11434"),
+            "http://[::1]:11434"
+        );
+    }
+
+    #[test]
+    fn normalize_preserves_named_host() {
+        assert_eq!(
+            normalize_unspecified_host("http://localhost:11434"),
+            "http://localhost:11434"
+        );
+        assert_eq!(
+            normalize_unspecified_host("http://my-ollama.lan:11434"),
+            "http://my-ollama.lan:11434"
+        );
+    }
+
+    #[test]
+    fn normalize_preserves_private_ip() {
+        assert_eq!(
+            normalize_unspecified_host("http://192.168.1.5:11434"),
+            "http://192.168.1.5:11434"
+        );
+    }
+
+    #[test]
+    fn normalize_handles_invalid_url() {
+        assert_eq!(normalize_unspecified_host("not a url"), "not a url");
+    }
+
+    // ── ollama_base_url: 0.0.0.0 normalization ─────────────────────────
+
+    #[test]
+    fn ollama_base_url_normalizes_unspecified_in_env_override() {
+        let _lock = test_lock();
+        let _g = OllamaEnvGuard::set("http://0.0.0.0:11434");
+        assert_eq!(ollama_base_url(), "http://localhost:11434");
+    }
+
+    #[test]
+    fn ollama_base_url_normalizes_unspecified_in_ollama_host() {
+        let _lock = test_lock();
+        let _g1 = OllamaEnvGuard::clear();
+        let _g2 = OllamaEnvGuard::set_var(OLLAMA_HOST_VAR, "0.0.0.0:11434");
+        assert_eq!(ollama_base_url(), "http://localhost:11434");
+    }
+
+    #[test]
+    fn ollama_base_url_normalizes_ipv6_unspecified_in_ollama_host() {
+        let _lock = test_lock();
+        let _g1 = OllamaEnvGuard::clear();
+        let _g2 = OllamaEnvGuard::set_var(OLLAMA_HOST_VAR, "http://[::]:11434");
+        assert_eq!(ollama_base_url(), "http://[::1]:11434");
+    }
+
+    // ── ollama_base_url_from_config: 0.0.0.0 normalization ──────────────
+
+    #[test]
+    fn ollama_base_url_from_config_normalizes_unspecified() {
+        let _lock = test_lock();
+        let _g = OllamaEnvGuard::clear();
+        let config = make_config_with_base_url(Some("http://0.0.0.0:11434"));
+        assert_eq!(
+            ollama_base_url_from_config(&config),
+            "http://localhost:11434"
+        );
+    }
+
+    #[test]
+    fn ollama_base_url_from_config_normalizes_ipv6_unspecified() {
+        let _lock = test_lock();
+        let _g = OllamaEnvGuard::clear();
+        let config = make_config_with_base_url(Some("http://[::]:11434"));
+        assert_eq!(ollama_base_url_from_config(&config), "http://[::1]:11434");
+    }
+
     // ── validate_ollama_url ───────────────────────────────────────────
 
     #[test]
@@ -734,6 +872,22 @@ mod tests {
         );
     }
 
+    #[test]
+    fn validate_ollama_url_rewrites_ipv4_unspecified_to_localhost() {
+        assert_eq!(
+            validate_ollama_url("http://0.0.0.0:11434"),
+            Ok("http://localhost:11434".to_string())
+        );
+    }
+
+    #[test]
+    fn validate_ollama_url_rewrites_ipv6_unspecified_to_loopback() {
+        assert_eq!(
+            validate_ollama_url("http://[::]:11434"),
+            Ok("http://[::1]:11434".to_string())
+        );
+    }
+
     // ── redact_ollama_base_url ────────────────────────────────────────
 
     #[test]
diff --git a/src/openhuman/inference/local/service/assets.rs b/src/openhuman/inference/local/service/assets.rs
index b45fbcf2e..2c224eed4 100644
--- a/src/openhuman/inference/local/service/assets.rs
+++ b/src/openhuman/inference/local/service/assets.rs
@@ -45,7 +45,8 @@ impl LocalAiService {
             LocalAiProvider::Ollama | LocalAiProvider::LmStudio
         );
         let ollama_available = if uses_ollama_assets {
-            let present = self.ollama_healthy().await;
+            let base_url = crate::openhuman::inference::local::ollama_base_url_from_config(config);
+            let present = self.ollama_healthy_at(&base_url).await;
             debug!(
                 target: "local_ai::assets",
                 %correlation_id,
@@ -89,7 +90,7 @@ impl LocalAiService {
                 }
             };
             let embedding_ready = if ollama_available {
-                match self.has_model(&embedding_model).await {
+                match self.has_model_for_config(config, &embedding_model).await {
                     Ok(ready) => {
                         debug!(
                             target: "local_ai::assets",
@@ -131,7 +132,7 @@ impl LocalAiService {
                 branch = "ollama",
                 "[local_ai:assets:provider_routing] selected provider branch"
             );
-            let chat_ready = match self.has_model(&chat_model).await {
+            let chat_ready = match self.has_model_for_config(config, &chat_model).await {
                 Ok(ready) => {
                     debug!(
                         target: "local_ai::assets",
@@ -157,7 +158,7 @@ impl LocalAiService {
                     false
                 }
             };
-            let vision_ready = match self.has_model(&vision_model).await {
+            let vision_ready = match self.has_model_for_config(config, &vision_model).await {
                 Ok(ready) => {
                     debug!(
                         target: "local_ai::assets",
@@ -183,7 +184,7 @@ impl LocalAiService {
                     false
                 }
             };
-            let embedding_ready = match self.has_model(&embedding_model).await {
+            let embedding_ready = match self.has_model_for_config(config, &embedding_model).await {
                 Ok(ready) => {
                     debug!(
                         target: "local_ai::assets",
@@ -533,7 +534,7 @@ impl LocalAiService {
                 }
                 if let Err(err) = async {
                     self.ensure_ollama_server(config).await?;
-                    self.ensure_ollama_model_available(&model_id, "embedding")
+                    self.ensure_ollama_model_available(config, &model_id, "embedding")
                         .await
                 }
                 .await
@@ -625,7 +626,8 @@ impl LocalAiService {
                     _ => {}
                 }
             }
-            self.ensure_ollama_model_available(&model_id, label).await?;
+            self.ensure_ollama_model_available(config, &model_id, label)
+                .await?;
         }
 
         let mut stt_warning = None;
@@ -687,7 +689,8 @@ impl LocalAiService {
             "chat" => {
                 self.ensure_ollama_server(config).await?;
                 let model = model_ids::effective_chat_model_id(config);
-                self.ensure_ollama_model_available(&model, "chat").await?;
+                self.ensure_ollama_model_available(config, &model, "chat")
+                    .await?;
             }
             "vision" => {
                 if matches!(
@@ -701,12 +704,13 @@ impl LocalAiService {
                 }
                 self.ensure_ollama_server(config).await?;
                 let model = model_ids::effective_vision_model_id(config);
-                self.ensure_ollama_model_available(&model, "vision").await?;
+                self.ensure_ollama_model_available(config, &model, "vision")
+                    .await?;
             }
             "embedding" | "embeddings" => {
                 self.ensure_ollama_server(config).await?;
                 let model = model_ids::effective_embedding_model_id(config);
-                self.ensure_ollama_model_available(&model, "embedding")
+                self.ensure_ollama_model_available(config, &model, "embedding")
                     .await?;
             }
             "stt" => {
diff --git a/src/openhuman/inference/local/service/bootstrap.rs b/src/openhuman/inference/local/service/bootstrap.rs
index 782eb9414..bb279708d 100644
--- a/src/openhuman/inference/local/service/bootstrap.rs
+++ b/src/openhuman/inference/local/service/bootstrap.rs
@@ -223,7 +223,7 @@ impl LocalAiService {
                     log::trace!(
                         "[local_ai] LM Studio bootstrap embedding ensure_ollama_model_available start model={embedding_model}"
                     );
-                    self.ensure_ollama_model_available(&embedding_model, "embedding")
+                    self.ensure_ollama_model_available(&effective_config, &embedding_model, "embedding")
                         .await?;
                     log::trace!(
                         "[local_ai] LM Studio bootstrap embedding ensure_ollama_model_available succeeded model={embedding_model}"
diff --git a/src/openhuman/inference/local/service/ollama_admin.rs b/src/openhuman/inference/local/service/ollama_admin.rs
index 7d0089c95..de2b0e760 100644
--- a/src/openhuman/inference/local/service/ollama_admin.rs
+++ b/src/openhuman/inference/local/service/ollama_admin.rs
@@ -541,7 +541,7 @@ impl LocalAiService {
         config: &Config,
     ) -> Result<(), String> {
         let chat_model = model_ids::effective_chat_model_id(config);
-        self.ensure_ollama_model_available(&chat_model, "chat")
+        self.ensure_ollama_model_available(config, &chat_model, "chat")
             .await?;
 
         match presets::vision_mode_for_config(&config.local_ai) {
@@ -553,7 +553,7 @@ impl LocalAiService {
             }
             VisionMode::Bundled => {
                 let vision_model = model_ids::effective_vision_model_id(config);
-                self.ensure_ollama_model_available(&vision_model, "vision")
+                self.ensure_ollama_model_available(config, &vision_model, "vision")
                     .await?;
                 self.status.lock().vision_state = "ready".to_string();
             }
@@ -561,7 +561,7 @@ impl LocalAiService {
 
         let embedding_model = model_ids::effective_embedding_model_id(config);
         if config.local_ai.preload_embedding_model {
-            self.ensure_ollama_model_available(&embedding_model, "embedding")
+            self.ensure_ollama_model_available(config, &embedding_model, "embedding")
                 .await?;
             self.status.lock().embedding_state = "ready".to_string();
         }
@@ -579,10 +579,12 @@ impl LocalAiService {
 
     pub(in crate::openhuman::inference::local::service) async fn ensure_ollama_model_available(
         &self,
+        config: &Config,
         model_id: &str,
         label: &str,
     ) -> Result<(), String> {
-        if self.has_model(model_id).await? {
+        let base_url = ollama_base_url_from_config(config);
+        if self.has_model_at(&base_url, model_id).await? {
             return Ok(());
         }
 
@@ -635,7 +637,7 @@ impl LocalAiService {
 
             let response = match self
                 .http
-                .post(format!("{}/api/pull", ollama_base_url()))
+                .post(format!("{base_url}/api/pull"))
                 .json(&OllamaPullRequest {
                     name: model_id.to_string(),
                     stream: true,
@@ -735,6 +737,7 @@ impl LocalAiService {
                 last_error = Some(err.clone());
                 let resumed = self
                     .wait_for_model_after_pull_interruption(
+                        &base_url,
                         model_id,
                         attempt,
                         MAX_PULL_RETRIES,
@@ -751,7 +754,7 @@ impl LocalAiService {
                 return Err(format!("{err} after {MAX_PULL_RETRIES} attempts"));
             }
 
-            if self.has_model(model_id).await? {
+            if self.has_model_at(&base_url, model_id).await? {
                 break;
             }
 
@@ -761,6 +764,7 @@ impl LocalAiService {
             ));
             let resumed = self
                 .wait_for_model_after_pull_interruption(
+                    &base_url,
                     model_id,
                     attempt,
                     MAX_PULL_RETRIES,
@@ -776,7 +780,7 @@ impl LocalAiService {
             }
         }
 
-        if !self.has_model(model_id).await? {
+        if !self.has_model_at(&base_url, model_id).await? {
             return Err(last_error.unwrap_or_else(|| {
                 format!(
                     "ollama pull finished but model `{}` was not found",
@@ -796,6 +800,7 @@ impl LocalAiService {
 
     async fn wait_for_model_after_pull_interruption(
         &self,
+        base_url: &str,
         model_id: &str,
         attempt: usize,
         max_attempts: usize,
@@ -826,7 +831,7 @@ impl LocalAiService {
 
         let deadline = std::time::Instant::now() + std::time::Duration::from_secs(wait_secs);
         while std::time::Instant::now() < deadline {
-            if self.has_model(model_id).await? {
+            if self.has_model_at(base_url, model_id).await? {
                 log::info!(
                     "[local_ai] model `{}` became available after interrupted pull stream",
                     model_id
@@ -1407,6 +1412,15 @@ impl LocalAiService {
         self.has_model_at(&ollama_base_url(), model).await
     }
 
+    pub(in crate::openhuman::inference::local::service) async fn has_model_for_config(
+        &self,
+        config: &Config,
+        model: &str,
+    ) -> Result<bool, String> {
+        self.has_model_at(&ollama_base_url_from_config(config), model)
+            .await
+    }
+
     async fn has_model_at(&self, base_url: &str, model: &str) -> Result<bool, String> {
         // Issue the /api/tags GET directly. We previously short-circuited via
         // ollama_healthy(), but that doubled the number of /api/tags round-trips
diff --git a/src/openhuman/inference/local/service/vision_embed.rs b/src/openhuman/inference/local/service/vision_embed.rs
index 724b5fb66..0983392c1 100644
--- a/src/openhuman/inference/local/service/vision_embed.rs
+++ b/src/openhuman/inference/local/service/vision_embed.rs
@@ -36,7 +36,7 @@ impl LocalAiService {
         }
         self.bootstrap(config).await;
         let vision_model = model_ids::effective_vision_model_id(config);
-        self.ensure_ollama_model_available(&vision_model, "vision")
+        self.ensure_ollama_model_available(config, &vision_model, "vision")
             .await?;
 
         let images: Vec<String> = image_refs
@@ -148,7 +148,7 @@ impl LocalAiService {
         }
         self.bootstrap(config).await;
         let embedding_model = model_ids::effective_embedding_model_id(config);
-        self.ensure_ollama_model_available(&embedding_model, "embedding")
+        self.ensure_ollama_model_available(config, &embedding_model, "embedding")
             .await?;
 
         // Embeds are bge-m3 calls (8K context, ~1.3 GB resident) — the
diff --git a/src/openhuman/inference/schemas.rs b/src/openhuman/inference/schemas.rs
index 895e10cb6..e275c9294 100644
--- a/src/openhuman/inference/schemas.rs
+++ b/src/openhuman/inference/schemas.rs
@@ -1,4 +1,4 @@
-use serde::de::DeserializeOwned;
+use serde::de::{DeserializeOwned, Deserializer};
 use serde::Deserialize;
 use serde_json::{Map, Value};
 
@@ -92,7 +92,8 @@ struct InferenceUpdateLocalSettingsParams {
     runtime_enabled: Option<bool>,
     opt_in_confirmed: Option<bool>,
     provider: Option<String>,
-    base_url: Option<String>,
+    #[serde(default, deserialize_with = "deserialize_present_json")]
+    base_url: Option<Value>,
     model_id: Option<String>,
     chat_model_id: Option<String>,
     usage_embeddings: Option<bool>,
@@ -276,7 +277,10 @@ pub fn schemas(function: &str) -> ControllerSchema {
                 optional_bool("runtime_enabled", "Enable or disable local inference runtime routing."),
                 optional_bool("opt_in_confirmed", "Persist the local inference opt-in flag."),
                 optional_string("provider", "Optional local provider slug, e.g. ollama or lm_studio."),
-                optional_string("base_url", "Optional local provider base URL."),
+                optional_json(
+                    "base_url",
+                    "Optional local provider base URL string, or null to clear.",
+                ),
                 optional_string("model_id", "Optional generic model id override."),
                 optional_string("chat_model_id", "Optional chat model id override."),
                 optional_bool("usage_embeddings", "Whether embeddings workload may use the local provider."),
@@ -629,11 +633,17 @@ fn handle_inference_update_model_settings(params: Map<String, Value>) -> Control
 fn handle_inference_update_local_settings(params: Map<String, Value>) -> ControllerFuture {
     Box::pin(async move {
         let update = deserialize_params::<InferenceUpdateLocalSettingsParams>(params)?;
+        let base_url = match update.base_url {
+            None => None,
+            Some(Value::Null) => Some(None),
+            Some(Value::String(value)) => Some(Some(value)),
+            Some(_) => return Err("invalid params: base_url must be a string or null".to_string()),
+        };
         let patch = config_rpc::LocalAiSettingsPatch {
             runtime_enabled: update.runtime_enabled,
             opt_in_confirmed: update.opt_in_confirmed,
             provider: update.provider,
-            base_url: update.base_url,
+            base_url,
             model_id: update.model_id,
             chat_model_id: update.chat_model_id,
             usage_embeddings: update.usage_embeddings,
@@ -812,6 +822,13 @@ fn deserialize_params<T: DeserializeOwned>(params: Map<String, Value>) -> Result
     serde_json::from_value(Value::Object(params)).map_err(|e| format!("invalid params: {e}"))
 }
 
+fn deserialize_present_json<'de, D>(deserializer: D) -> Result<Option<Value>, D::Error>
+where
+    D: Deserializer<'de>,
+{
+    Value::deserialize(deserializer).map(Some)
+}
+
 fn to_json<T: serde::Serialize>(outcome: RpcOutcome<T>) -> Result<Value, String> {
     outcome.into_cli_compatible_json()
 }
diff --git a/src/openhuman/inference/schemas_tests.rs b/src/openhuman/inference/schemas_tests.rs
index 1370ef929..38e83b40f 100644
--- a/src/openhuman/inference/schemas_tests.rs
+++ b/src/openhuman/inference/schemas_tests.rs
@@ -58,6 +58,20 @@ fn inference_prompt_schema_reuses_local_ai_shape_with_new_namespace() {
     assert!(schema.inputs.iter().any(|field| field.name == "max_tokens"));
 }
 
+#[test]
+fn inference_update_local_settings_schema_allows_json_base_url() {
+    let schema = schemas("update_local_settings");
+    let field = schema
+        .inputs
+        .iter()
+        .find(|field| field.name == "base_url")
+        .expect("base_url field");
+    match &field.ty {
+        TypeSchema::Option(inner) => assert!(matches!(**inner, TypeSchema::Json)),
+        other => panic!("expected Option<Json>, got {other:?}"),
+    }
+}
+
 #[test]
 fn inference_openai_oauth_schemas_are_registered_with_expected_shapes() {
     let registered: Vec<&str> = all_registered_controllers()
diff --git a/tests/json_rpc_e2e.rs b/tests/json_rpc_e2e.rs
index 539a82737..afb49fe8a 100644
--- a/tests/json_rpc_e2e.rs
+++ b/tests/json_rpc_e2e.rs
@@ -5582,6 +5582,66 @@ async fn json_rpc_local_ai_lm_studio_config_diagnostics_and_prompt() {
     rpc_join.abort();
 }
 
+#[tokio::test]
+async fn json_rpc_local_ai_ollama_endpoint_normalizes_bind_address_and_clears() {
+    let _env_lock = json_rpc_e2e_env_lock();
+    let tmp = tempdir().expect("tempdir");
+    let home = tmp.path();
+    let openhuman_home = home.join(".openhuman");
+
+    let _home_guard = EnvVarGuard::set_to_path("HOME", home);
+    let _workspace_guard = EnvVarGuard::unset("OPENHUMAN_WORKSPACE");
+    let _backend_url_guard = EnvVarGuard::unset("BACKEND_URL");
+    let _vite_backend_guard = EnvVarGuard::unset("VITE_BACKEND_URL");
+
+    let (mock_addr, mock_join) = serve_on_ephemeral(mock_upstream_router()).await;
+    let mock_origin = format!("http://{}", mock_addr);
+    write_min_config(&openhuman_home, &mock_origin);
+
+    let (rpc_addr, rpc_join) = serve_on_ephemeral(build_core_http_router(false)).await;
+    let rpc_base = format!("http://{}", rpc_addr);
+    tokio::time::sleep(Duration::from_millis(100)).await;
+
+    let update = post_json_rpc(
+        &rpc_base,
+        390,
+        "openhuman.inference_update_local_settings",
+        json!({
+            "provider": "ollama",
+            "base_url": "http://0.0.0.0:11434/api/tags"
+        }),
+    )
+    .await;
+    let update_result = assert_no_jsonrpc_error(&update, "normalize ollama base_url");
+    assert_eq!(
+        update_result
+            .get("result")
+            .and_then(|value| value.get("config"))
+            .and_then(|config| config.get("local_ai"))
+            .and_then(|local_ai| local_ai.get("base_url"))
+            .and_then(Value::as_str),
+        Some("http://localhost:11434")
+    );
+
+    let clear = post_json_rpc(
+        &rpc_base,
+        391,
+        "openhuman.inference_update_local_settings",
+        json!({ "base_url": null }),
+    )
+    .await;
+    let clear_result = assert_no_jsonrpc_error(&clear, "clear ollama base_url");
+    assert!(clear_result
+        .get("result")
+        .and_then(|value| value.get("config"))
+        .and_then(|config| config.get("local_ai"))
+        .and_then(|local_ai| local_ai.get("base_url"))
+        .is_none_or(Value::is_null));
+
+    mock_join.abort();
+    rpc_join.abort();
+}
+
 #[tokio::test]
 async fn json_rpc_inference_namespace_lm_studio_prompt_and_status() {
     let _env_lock = json_rpc_e2e_env_lock();
diff --git a/tests/memory_sync_sources_raw_coverage_e2e.rs b/tests/memory_sync_sources_raw_coverage_e2e.rs
index c195bfde4..7b803035f 100644
--- a/tests/memory_sync_sources_raw_coverage_e2e.rs
+++ b/tests/memory_sync_sources_raw_coverage_e2e.rs
@@ -404,6 +404,9 @@ async fn github_reader_uses_fake_gh_for_list_and_read_paths() {
     let reader = openhuman_core::openhuman::memory_sources::readers::github::GithubReader;
     let mut entry = source(SourceKind::GithubRepo, "github-round15");
     entry.url = Some("https://github.com/tinyhumansai/openhuman.git".to_string());
+    entry.max_commits = Some(30);
+    entry.max_issues = Some(30);
+    entry.max_prs = Some(30);
 
     let items = reader
         .list_items(&entry, &config)