diff --git a/src/chrome/src/providers/manager.js b/src/chrome/src/providers/manager.js
index 984f1910a..1a8842e1f 100644
--- a/src/chrome/src/providers/manager.js
+++ b/src/chrome/src/providers/manager.js
@@ -274,6 +274,7 @@ export class ProviderManager {
         category: 'local',
         label: 'Ollama (Local)',
         providerName: 'ollama',
+        requiresModel: true,
         baseUrl: 'http://localhost:11434/v1',
         model: '',
         contextWindow: 16384,
@@ -302,6 +303,7 @@ export class ProviderManager {
         category: 'local',
         label: 'Jan (Local)',
         providerName: 'jan',
+        requiresModel: true,
         baseUrl: 'http://localhost:1337/v1',
         model: '',
         contextWindow: 16384,
@@ -315,6 +317,7 @@ export class ProviderManager {
         category: 'local',
         label: 'vLLM (Local)',
         providerName: 'vllm',
+        requiresModel: true,
         baseUrl: 'http://localhost:8000/v1',
         model: '',
         contextWindow: 16384,
@@ -328,6 +331,7 @@ export class ProviderManager {
         category: 'local',
         label: 'SGLang (Local)',
         providerName: 'sglang',
+        requiresModel: true,
         baseUrl: 'http://localhost:30000/v1',
         model: '',
         contextWindow: 16384,
@@ -341,6 +345,7 @@ export class ProviderManager {
         category: 'local',
         label: 'LocalAI (Local)',
         providerName: 'localai',
+        requiresModel: true,
         baseUrl: 'http://localhost:8080/v1',
         model: '',
         contextWindow: 16384,
@@ -355,6 +360,7 @@ export class ProviderManager {
         category: 'local',
         label: 'GPT4All (Local)',
         providerName: 'gpt4all',
+        requiresModel: true,
         baseUrl: 'http://localhost:4891/v1',
         model: '',
         contextWindow: 16384,
diff --git a/src/chrome/src/providers/openai.js b/src/chrome/src/providers/openai.js
index 95804b3ed..72651d20c 100644
--- a/src/chrome/src/providers/openai.js
+++ b/src/chrome/src/providers/openai.js
@@ -79,10 +79,10 @@ export class OpenAICompatibleProvider extends BaseLLMProvider {
   get model() {
     if (this.config.model) return this.config.model;
     if (this.config.requiresModel) throw new Error(`${this.config.label || this.name} model is required.`);
-    // Local servers (Ollama, LM Studio, vLLM, …) must never receive a model
-    // id the user never configured: most 404 on unknown ids and none serves
-    // a model named after OpenAI's default. Omit the field entirely so the
-    // server applies its own default. Mirrors LlamaCppProvider.
+    // Some local servers apply their own default when no model is configured.
+    // Others carry `requiresModel: true` and throw above. Treat the category as
+    // the durable boundary so older/custom local entries never inherit a
+    // fabricated cloud model id merely because their provider name is unknown.
     if (this.config.category === 'local') return null;
     return String(this.config.providerName || '').toLowerCase() === 'openai'
       && this._isOfficialOpenAIBaseUrl()
diff --git a/src/firefox/src/providers/manager.js b/src/firefox/src/providers/manager.js
index 0bcfdf6a3..73b6e9740 100644
--- a/src/firefox/src/providers/manager.js
+++ b/src/firefox/src/providers/manager.js
@@ -247,6 +247,7 @@ export class ProviderManager {
         category: 'local',
         label: 'Ollama (Local)',
         providerName: 'ollama',
+        requiresModel: true,
         baseUrl: 'http://localhost:11434/v1',
         model: '',
         contextWindow: 16384,
@@ -275,6 +276,7 @@ export class ProviderManager {
         category: 'local',
         label: 'Jan (Local)',
         providerName: 'jan',
+        requiresModel: true,
         baseUrl: 'http://localhost:1337/v1',
         model: '',
         contextWindow: 16384,
@@ -288,6 +290,7 @@ export class ProviderManager {
         category: 'local',
         label: 'vLLM (Local)',
         providerName: 'vllm',
+        requiresModel: true,
         baseUrl: 'http://localhost:8000/v1',
         model: '',
         contextWindow: 16384,
@@ -301,6 +304,7 @@ export class ProviderManager {
         category: 'local',
         label: 'SGLang (Local)',
         providerName: 'sglang',
+        requiresModel: true,
         baseUrl: 'http://localhost:30000/v1',
         model: '',
         contextWindow: 16384,
@@ -314,6 +318,7 @@ export class ProviderManager {
         category: 'local',
         label: 'LocalAI (Local)',
         providerName: 'localai',
+        requiresModel: true,
         baseUrl: 'http://localhost:8080/v1',
         model: '',
         contextWindow: 16384,
@@ -328,6 +333,7 @@ export class ProviderManager {
         category: 'local',
         label: 'GPT4All (Local)',
         providerName: 'gpt4all',
+        requiresModel: true,
         baseUrl: 'http://localhost:4891/v1',
         model: '',
         contextWindow: 16384,
diff --git a/src/firefox/src/providers/openai.js b/src/firefox/src/providers/openai.js
index 26458236d..8fa039933 100644
--- a/src/firefox/src/providers/openai.js
+++ b/src/firefox/src/providers/openai.js
@@ -79,10 +79,10 @@ export class OpenAICompatibleProvider extends BaseLLMProvider {
   get model() {
     if (this.config.model) return this.config.model;
     if (this.config.requiresModel) throw new Error(`${this.config.label || this.name} model is required.`);
-    // Local servers (Ollama, LM Studio, vLLM, …) must never receive a model
-    // id the user never configured: most 404 on unknown ids and none serves
-    // a model named after OpenAI's default. Omit the field entirely so the
-    // server applies its own default. Mirrors LlamaCppProvider.
+    // Some local servers apply their own default when no model is configured.
+    // Others carry `requiresModel: true` and throw above. Treat the category as
+    // the durable boundary so older/custom local entries never inherit a
+    // fabricated cloud model id merely because their provider name is unknown.
     if (this.config.category === 'local') return null;
     return String(this.config.providerName || '').toLowerCase() === 'openai'
       && this._isOfficialOpenAIBaseUrl()
diff --git a/test/run.js b/test/run.js
index 6894e9e6f..53a10937d 100644
--- a/test/run.js
+++ b/test/run.js
@@ -49686,6 +49686,66 @@ test('official OpenAI GPT-5.6 and Responses-only GPT-5 Pro variants route to Res
   }
 });
 
+test('local OpenAI-compatible servers that require a model throw a clear error when unset', () => {
+  for (const Provider of [OpenAIProviderCh, OpenAIProviderFx]) {
+    for (const providerName of ['ollama', 'jan', 'vllm', 'sglang', 'localai', 'gpt4all', 'local_openai_proxy']) {
+      const provider = new Provider({
+        providerName,
+        category: 'local',
+        baseUrl: 'http://localhost:1234/v1',
+        requiresModel: true,
+      });
+      assert.throws(
+        () => provider._buildChatCompletionsBody([{ role: 'user', content: 'hello' }], {}),
+        /model is required/,
+        `${providerName}: empty model must fail loudly instead of sending a fabricated id`,
+      );
+    }
+  }
+});
+
+test('every optional-model local provider omits the model field when unset', () => {
+  for (const Provider of [OpenAIProviderCh, OpenAIProviderFx]) {
+    for (const providerName of ['lmstudio', 'privatemode-ai', 'persisted-custom-local']) {
+      const empty = new Provider({
+        providerName,
+        category: 'local',
+        baseUrl: 'http://localhost:1234/v1',
+      });
+      assert.equal(empty.model, null, `${providerName}: local providers must not fabricate a model id`);
+      const chatBody = empty._buildChatCompletionsBody([{ role: 'user', content: 'hello' }], {});
+      assert.equal('model' in chatBody, false, `${providerName}: Chat Completions must omit an unset model`);
+      const responsesBody = empty._responsesBody([{ role: 'user', content: 'hello' }], {}, false);
+      assert.equal('model' in responsesBody, false, `${providerName}: Responses must omit an unset model`);
+    }
+
+    const configured = new Provider({
+      providerName: 'persisted-custom-local',
+      category: 'local',
+      baseUrl: 'http://localhost:1234/v1',
+      model: 'llama-3.2-3b',
+    });
+    assert.equal(
+      configured._buildChatCompletionsBody([{ role: 'user', content: 'hello' }], {}).model,
+      'llama-3.2-3b',
+      'configured local model must be sent',
+    );
+  }
+});
+
+test('non-local OpenAI-compatible providers keep the legacy model fallback', () => {
+  for (const Provider of [OpenAIProviderCh, OpenAIProviderFx]) {
+    assert.equal(new Provider({ providerName: 'openrouter' }).model, 'gpt-4o');
+    assert.equal(new Provider({ providerName: 'openai', baseUrl: 'https://proxy.example.test/v1' }).model, 'gpt-4o');
+    assert.equal(new Provider({ providerName: 'openai' }).model, 'gpt-5.6-terra');
+    const body = new Provider({ providerName: 'openrouter' })._buildChatCompletionsBody(
+      [{ role: 'user', content: 'hello' }],
+      {},
+    );
+    assert.equal(body.model, 'gpt-4o', 'non-local fallback model must stay on the wire');
+  }
+});
+
 test('Responses reasoning effort is normalized for GPT-5 Pro model constraints', () => {
   for (const Provider of [OpenAIProviderCh, OpenAIProviderFx]) {
     for (const model of ['gpt-5-pro', 'gpt-5-pro-2025-10-06']) {