fix(core): honor Retry-After for model retries

packages/core/src/agent/agent.spec.ts

@@ -3689,6 +3689,82 @@ Use pandas and summarize findings.`.split("\n"),
       }
     });
 
+    it("should honor Retry-After when retrying provider rate limits", async () => {
+      vi.useFakeTimers();
+      const setTimeoutSpy = vi.spyOn(globalThis, "setTimeout");
+
+      let resolveRetry!: () => void;
+      const retrySeen = new Promise<void>((resolve) => {
+        resolveRetry = resolve;
+      });
+      const onRetry = vi.fn(() => {
+        resolveRetry();
+      });
+      const agent = new Agent({
+        name: "RetryAfterAgent",
+        instructions: "Test",
+        model: mockModel as any,
+        maxRetries: 1,
+        hooks: createHooks({ onRetry }),
+      });
+
+      const mockResponse = {
+        text: "Retry response",
+        content: [{ type: "text", text: "Retry response" }],
+        reasoning: [],
+        files: [],
+        sources: [],
+        toolCalls: [],
+        toolResults: [],
+        finishReason: "stop",
+        usage: {
+          inputTokens: 10,
+          outputTokens: 5,
+          totalTokens: 15,
+        },
+        warnings: [],
+        request: {},
+        response: {
+          id: "retry-response",
+          modelId: "test-model",
+          timestamp: new Date(),
+          messages: [],
+        },
+        steps: [],
+      };
+
+      let callCount = 0;
+      vi.mocked(ai.generateText).mockImplementation(async () => {
+        callCount += 1;
+        if (callCount === 1) {
+          const error = new Error("Rate limited");
+          (error as any).isRetryable = true;
+          (error as any).statusCode = 429;
+          (error as any).headers = new Headers({ "retry-after": "3" });
+          throw error;
+        }
+        return mockResponse as any;
+      });
+
+      const resultPromise = agent.generateText("Test");
+
+      try {
+        await retrySeen;
+        await Promise.resolve();
+
+        expect(onRetry).toHaveBeenCalledTimes(1);
+        expect(vi.mocked(ai.generateText)).toHaveBeenCalledTimes(1);
+        expect(setTimeoutSpy).toHaveBeenCalledWith(expect.any(Function), 3000);
+
+        await vi.advanceTimersByTimeAsync(3000);
+        await expect(resultPromise).resolves.toMatchObject({ text: "Retry response" });
+        expect(vi.mocked(ai.generateText)).toHaveBeenCalledTimes(2);
+      } finally {
+        setTimeoutSpy.mockRestore();
+        vi.useRealTimers();
+      }
+    });
+
     it("should handle model errors gracefully", async () => {
       const agent = new Agent({
         name: "TestAgent",

packages/core/src/agent/agent.ts

@@ -5738,6 +5738,30 @@ export class Agent {
     return true;
   }
 
+  private getRetryAfterDelayMs(error: unknown): number | undefined {
+    const headers = (error as { headers?: Headers | Record<string, string> } | undefined)?.headers;
+    const retryAfter =
+      headers instanceof Headers
+        ? headers.get("retry-after")
+        : (headers?.["retry-after"] ?? headers?.["Retry-After"]);
+
+    if (!retryAfter) {
+      return undefined;
+    }
+
+    const seconds = Number.parseInt(retryAfter, 10);
+    if (Number.isFinite(seconds) && seconds > 0) {
+      return seconds * 1000;
+    }
+
+    const retryAt = Date.parse(retryAfter);
+    if (Number.isFinite(retryAt)) {
+      return Math.max(retryAt - Date.now(), 0);
+    }
+
+    return undefined;
+  }
+
   private async executeWithModelFallback<T>({
     oc,
     operation,
@@ -5885,7 +5909,8 @@ export class Agent {
           const canRetry = retryEligible && !isLastAttempt;
 
           if (canRetry) {
-            const retryDelayMs = Math.min(1000 * 2 ** attemptIndex, 10000);
+            const retryDelayMs =
+              this.getRetryAfterDelayMs(error) ?? Math.min(1000 * 2 ** attemptIndex, 10000);
             logger.debug(`[Agent:${this.name}] - Model attempt failed, retrying`, {
               operation,
               modelName,