fix(core): skip stream_options for providers that reject it (Cerebras 422)

packages/core/src/core/openaiContentGenerator.test.ts

@@ -388,9 +388,73 @@ describe('OpenAIContentGenerator', () => {
           temperature: 0.7, // From config sampling params (higher priority)
           max_tokens: 1000, // From config sampling params (higher priority)
           top_p: 0.9,
-        }),
+          }),
       );
     });
+
+    it('should omit store for strict providers like Cerebras', async () => {
+      vi.stubEnv('OPENAI_BASE_URL', 'https://api.cerebras.ai/v1');
+      generator = new OpenAIContentGenerator('test-key', 'gpt-4', mockConfig);
+      mockOpenAIClient.baseURL = 'https://api.cerebras.ai/v1';
+
+      const mockResponse = {
+        id: 'chatcmpl-123',
+        choices: [
+          {
+            index: 0,
+            message: { role: 'assistant', content: 'Response' },
+            finish_reason: 'stop',
+          },
+        ],
+        created: 1677652288,
+        model: 'gpt-4',
+      };
+      mockOpenAIClient.chat.completions.create.mockResolvedValue(mockResponse);
+
+      const request: GenerateContentParameters = {
+        contents: [{ role: 'user', parts: [{ text: 'Hello' }] }],
+        model: 'gpt-4',
+      };
+
+      await generator.generateContent(request, 'test-prompt-id');
+
+      const createCall =
+        mockOpenAIClient.chat.completions.create.mock.calls[0]?.[0];
+      expect(createCall).toBeDefined();
+      expect(createCall).not.toHaveProperty('store');
+    });
+
+    it('should include store for regular OpenAI providers on GPT models', async () => {
+      vi.stubEnv('OPENAI_BASE_URL', 'https://api.openai.com/v1');
+      generator = new OpenAIContentGenerator('test-key', 'gpt-4', mockConfig);
+      mockOpenAIClient.baseURL = 'https://api.openai.com/v1';
+
+      const mockResponse = {
+        id: 'chatcmpl-123',
+        choices: [
+          {
+            index: 0,
+            message: { role: 'assistant', content: 'Response' },
+            finish_reason: 'stop',
+          },
+        ],
+        created: 1677652288,
+        model: 'gpt-4',
+      };
+      mockOpenAIClient.chat.completions.create.mockResolvedValue(mockResponse);
+
+      const request: GenerateContentParameters = {
+        contents: [{ role: 'user', parts: [{ text: 'Hello' }] }],
+        model: 'gpt-4',
+      };
+
+      await generator.generateContent(request, 'test-prompt-id');
+
+      const createCall =
+        mockOpenAIClient.chat.completions.create.mock.calls[0]?.[0];
+      expect(createCall).toBeDefined();
+      expect(createCall?.store).toBe(true);
+    });
   });
 
   describe('generateContentStream', () => {
@@ -570,6 +634,89 @@ describe('OpenAIContentGenerator', () => {
         }
       }
     });
+
+    it('should omit stream_options and store for strict providers like Cerebras', async () => {
+      vi.stubEnv('OPENAI_BASE_URL', 'https://api.cerebras.ai/v1');
+      generator = new OpenAIContentGenerator('test-key', 'gpt-4', mockConfig);
+      mockOpenAIClient.baseURL = 'https://api.cerebras.ai/v1';
+
+      mockOpenAIClient.chat.completions.create.mockResolvedValue({
+        async *[Symbol.asyncIterator]() {
+          yield {
+            id: 'chatcmpl-123',
+            choices: [
+              {
+                index: 0,
+                delta: { content: 'ok' },
+                finish_reason: 'stop',
+              },
+            ],
+            created: 1677652288,
+          };
+        },
+      });
+
+      const request: GenerateContentParameters = {
+        contents: [{ role: 'user', parts: [{ text: 'Hello' }] }],
+        model: 'gpt-4',
+      };
+
+      const stream = await generator.generateContentStream(
+        request,
+        'test-prompt-id',
+      );
+      for await (const _response of stream) {
+        // Exhaust stream
+      }
+
+      const createCall =
+        mockOpenAIClient.chat.completions.create.mock.calls[0]?.[0];
+      expect(createCall).toBeDefined();
+      expect(createCall).not.toHaveProperty('stream_options');
+      expect(createCall).not.toHaveProperty('store');
+    });
+
+    it('should include stream_options and store for regular OpenAI providers', async () => {
+      vi.stubEnv('OPENAI_BASE_URL', 'https://api.openai.com/v1');
+      generator = new OpenAIContentGenerator('test-key', 'gpt-4', mockConfig);
+      mockOpenAIClient.baseURL = 'https://api.openai.com/v1';
+
+      mockOpenAIClient.chat.completions.create.mockResolvedValue({
+        async *[Symbol.asyncIterator]() {
+          yield {
+            id: 'chatcmpl-123',
+            choices: [
+              {
+                index: 0,
+                delta: { content: 'ok' },
+                finish_reason: 'stop',
+              },
+            ],
+            created: 1677652288,
+          };
+        },
+      });
+
+      const request: GenerateContentParameters = {
+        contents: [{ role: 'user', parts: [{ text: 'Hello' }] }],
+        model: 'gpt-4',
+      };
+
+      const stream = await generator.generateContentStream(
+        request,
+        'test-prompt-id',
+      );
+      for await (const _response of stream) {
+        // Exhaust stream
+      }
+
+      const createCall =
+        mockOpenAIClient.chat.completions.create.mock.calls[0]?.[0];
+      expect(createCall).toBeDefined();
+      expect(createCall).toHaveProperty('stream_options');
+      expect(createCall?.stream_options).toEqual({ include_usage: true });
+      expect(createCall?.store).toBe(true);
+    });
   });
 
   describe('countTokens', () => {

packages/core/src/core/openaiContentGenerator.ts

@@ -164,6 +164,47 @@ export class OpenAIContentGenerator implements ContentGenerator {
     return false; // Default behavior: never suppress error logging
   }
 
+  /**
+   * Check if stream_options should be included in streaming requests.
+   * Some providers (e.g. Cerebras) strictly validate request parameters
+   * and return 422 for unknown fields like stream_options.
+   * Default: include stream_options (most OpenAI-compatible providers support it).
+   */
+  private shouldIncludeStreamOptions(): boolean {
+    const baseURL = this.client?.baseURL || '';
+    let hostname: string | undefined;
+    try {
+      hostname = new URL(baseURL).hostname;
+    } catch (_e) {
+      return true; // Default to including stream_options
+    }
+    // Providers known to reject stream_options with 422
+    const strictProviders = ['api.cerebras.ai'];
+    return !strictProviders.some(
+      (h) => hostname === h || hostname!.endsWith('.' + h),
+    );
+  }
+
+  /**
+   * Check if store should be included in requests.
+   * Some providers (e.g. Cerebras) reject unsupported fields like store with 422.
+   * Default: include store for GPT models unless provider is known strict.
+   */
+  private shouldIncludeStore(): boolean {
+    const baseURL = this.client?.baseURL || '';
+    let hostname: string | undefined;
+    try {
+      hostname = new URL(baseURL).hostname;
+    } catch (_e) {
+      return true; // Default to including store
+    }
+    // Providers known to reject store with 422
+    const strictProviders = ['api.cerebras.ai'];
+    return !strictProviders.some(
+      (h) => hostname === h || hostname!.endsWith('.' + h),
+    );
+  }
+
   /**
    * Check if metadata should be included in the request
    * Only include metadata for specific providers that support it
@@ -264,7 +305,9 @@ export class OpenAIContentGenerator implements ContentGenerator {
         modelName.includes('gpt5') ||
         modelName.includes('gpt4')
       ) {
-        createParams.store = true;
+        if (this.shouldIncludeStore()) {
+          createParams.store = true;
+        }
       }
 
       // Handle JSON schema requests (for generateJson calls)
@@ -431,7 +474,9 @@ export class OpenAIContentGenerator implements ContentGenerator {
         messages,
         ...samplingParams,
         stream: true,
-        stream_options: { include_usage: true },
+        ...(this.shouldIncludeStreamOptions() && {
+          stream_options: { include_usage: true },
+        }),
         ...(metadata && { metadata }),
       };
 
@@ -442,7 +487,9 @@ export class OpenAIContentGenerator implements ContentGenerator {
         modelNameStream.includes('gpt5') ||
         modelNameStream.includes('gpt4')
       ) {
-        createParams.store = true;
+        if (this.shouldIncludeStore()) {
+          createParams.store = true;
+        }
       }
 
       // Handle JSON schema requests (for generateJson calls) - same as non-streaming