feat: adapt cloudflare worker ai (#14732)

#### PR Dependency Tree


* **PR #14732** 👈

This tree was auto-generated by
[Charcoal](https://github.com/danerwilliams/charcoal)

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit

* **New Features**
* Cloudflare Workers AI added as a Copilot provider (configurable in
admin settings).
  * Gemini 3.1 Flash Lite Preview model made available.

* **Behavior Changes**
* Reranking now uses a different default model identifier (affects
relevancy scores).

* **Tests**
* Rerank tests adjusted to focus on the updated model and expected
results.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

packages/backend/server/src/__tests__/copilot/copilot-provider.spec.ts

@@ -964,11 +964,11 @@ test(
         'The stock market is experiencing significant fluctuations.',
       ];
 
-      const provider = (await factory.getProviderByModel('gpt-5.2'))!;
+      const provider = (await factory.getProviderByModel('gpt-4o-mini'))!;
       t.assert(provider, 'should have provider for rerank');
 
       const scores = await provider.rerank(
-        { modelId: 'gpt-5.2' },
+        { modelId: 'gpt-4o-mini' },
         {
           query,
           candidates: embeddings.map((text, index) => ({
@@ -990,8 +990,8 @@ test(
       t.log('Rerank scores:', scores);
       t.is(
         scores.filter(s => s > 0.5).length,
-        5,
-        'should have 5 related chunks'
+        4,
+        'should have 4 related chunks'
       );
     });
   }

packages/backend/server/src/__tests__/copilot/provider-native.spec.ts

@@ -3,6 +3,7 @@ import test from 'ava';
 
 import type { NativeLlmRerankRequest } from '../../native';
 import { ProviderMiddlewareConfig } from '../../plugins/copilot/config';
+import { CloudflareWorkersAIProvider } from '../../plugins/copilot/providers/cloudflare';
 import {
   normalizeOpenAIOptionsForModel,
   OpenAIProvider,
@@ -53,7 +54,7 @@ class TestOpenAIProvider extends CopilotProvider<{ apiKey: string }> {
 class NativeRerankProtocolProvider extends OpenAIProvider {
   override readonly models = [
     {
-      id: 'gpt-5.2',
+      id: 'gpt-4o-mini',
       capabilities: [
         {
           input: [ModelInputType.Text],
@@ -77,6 +78,19 @@ class NativeRerankProtocolProvider extends OpenAIProvider {
   }
 }
 
+class NativeCloudflareRerankProtocolProvider extends CloudflareWorkersAIProvider {
+  override get config() {
+    return {
+      apiToken: 'test-key',
+      accountId: 'account-1',
+    };
+  }
+
+  override configured() {
+    return true;
+  }
+}
+
 function createProvider(profileMiddleware?: ProviderMiddlewareConfig) {
   const provider = new TestOpenAIProvider();
   (provider as any).AFFiNEConfig = {
@@ -170,14 +184,14 @@ test('OpenAI rerank should always use chat-completions native protocol', async t
   ) => {
     capturedProtocol = protocol;
     capturedRequest = JSON.parse(requestJson) as NativeLlmRerankRequest;
-    return JSON.stringify({ model: 'gpt-5.2', scores: [0.9, 0.1] });
+    return JSON.stringify({ model: 'gpt-4o-mini', scores: [0.9, 0.1] });
   };
   t.teardown(() => {
     (serverNativeModule as any).llmRerankDispatch = original;
   });
 
   const scores = await provider.rerank(
-    { modelId: 'gpt-5.2' },
+    { modelId: 'gpt-4o-mini' },
     {
       query: 'programming',
       candidates: [
@@ -190,7 +204,7 @@ test('OpenAI rerank should always use chat-completions native protocol', async t
   t.deepEqual(scores, [0.9, 0.1]);
   t.is(capturedProtocol, 'openai_chat');
   t.deepEqual(capturedRequest, {
-    model: 'gpt-5.2',
+    model: 'gpt-4o-mini',
     query: 'programming',
     candidates: [
       { id: 'react', text: 'React is a UI library.' },
@@ -198,3 +212,63 @@ test('OpenAI rerank should always use chat-completions native protocol', async t
     ],
   });
 });
+
+test('Cloudflare rerank should keep native protocol details behind provider', async t => {
+  const provider = new NativeCloudflareRerankProtocolProvider();
+  let capturedProtocol: string | undefined;
+  let capturedRequest: NativeLlmRerankRequest | undefined;
+  let capturedBackendConfig: Record<string, unknown> | undefined;
+
+  const original = (serverNativeModule as any).llmRerankDispatch;
+  (serverNativeModule as any).llmRerankDispatch = (
+    protocol: string,
+    backendConfigJson: string,
+    requestJson: string
+  ) => {
+    capturedProtocol = protocol;
+    capturedBackendConfig = JSON.parse(backendConfigJson) as Record<
+      string,
+      unknown
+    >;
+    capturedRequest = JSON.parse(requestJson) as NativeLlmRerankRequest;
+    return JSON.stringify({
+      model: '@cf/qwen/qwen3-30b-a3b-fp8',
+      scores: [0.9, 0.1],
+    });
+  };
+  t.teardown(() => {
+    (serverNativeModule as any).llmRerankDispatch = original;
+  });
+
+  for (const modelId of [
+    '@cf/qwen/qwen3-30b-a3b-fp8',
+    '@cf/baai/bge-reranker-base',
+  ]) {
+    const scores = await provider.rerank(
+      { modelId },
+      {
+        query: 'programming',
+        candidates: [
+          { id: 'react', text: 'React is a UI library.' },
+          { id: 'weather', text: 'The weather is sunny today.' },
+        ],
+      }
+    );
+
+    t.deepEqual(scores, [0.9, 0.1]);
+    t.is(capturedProtocol, 'openai_chat');
+    t.deepEqual(capturedBackendConfig, {
+      base_url: 'https://api.cloudflare.com/client/v4/accounts/account-1/ai',
+      auth_token: 'test-key',
+      request_layer: 'cloudflare_workers_ai',
+    });
+    t.deepEqual(capturedRequest, {
+      model: modelId,
+      query: 'programming',
+      candidates: [
+        { id: 'react', text: 'React is a UI library.' },
+        { id: 'weather', text: 'The weather is sunny today.' },
+      ],
+    });
+  }
+});