fix: allow query strings and fragments in image URL regex (#484) (#485)

The supportedUrls['image/*'] regex anchored the extension to the end of
the string, so URLs with ?query or #fragment did not match and got
downloaded + base64-inlined by the AI SDK layer. Updated both chat and
completion models to accept an optional [?#].* suffix after the
extension. Strict allowlist behavior (extension required, terminal) is
preserved.

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

Author: robert-j-y

.changeset/fix-issue-484-image-url-query-params.md

@@ -0,0 +1,5 @@
+---
+'@openrouter/ai-sdk-provider': patch
+---
+
+Fix `supportedUrls['image/*']` regex to accept image URLs with query strings or fragments (e.g. `https://cdn.example.com/photo.png?height=200`, `.../photo.webp#frag`). Previously the `$` anchor on the extension caused such URLs to be treated as unsupported, forcing the AI SDK runtime to download and base64-inline them, which bloated conversation history and inflated token usage.

e2e/issues/issue-484-image-url-query-params.test.ts

@@ -0,0 +1,165 @@
+/**
+ * Regression test for GitHub issue #484
+ * https://github.com/OpenRouterTeam/ai-sdk-provider/issues/484
+ *
+ * Issue: "Zod Schema for Image URLs restricts query params"
+ *
+ * The `supportedUrls` regex for `image/*` only matches URLs that END with a
+ * known extension. Valid image URLs with query strings or fragments do NOT
+ * match, so the AI SDK's LanguageModelV3 layer treats them as unsupported
+ * and downloads + base64-encodes them.
+ */
+import { describe, expect, it } from 'vitest';
+import { createOpenRouter } from '@/src';
+
+const provider = createOpenRouter({ apiKey: 'test-key' });
+const chatModel = provider.chat('anthropic/claude-3.5-sonnet');
+const completionModel = provider.completion('openai/gpt-3.5-turbo-instruct');
+
+const IMAGE_URLS_WITH_QUERY_PARAMS = [
+  'https://cdn.example.com/photo.png?height=200',
+  'https://cdn.example.com/photo.jpeg?w=100&h=200',
+  'https://cdn.example.com/photo.webp?v=1&t=2',
+  'https://cdn.example.com/photo.gif?cache=false',
+  'https://cdn.example.com/photo.jpg?signature=abc123',
+] as const;
+
+const IMAGE_URLS_WITH_FRAGMENTS = [
+  'https://cdn.example.com/photo.png#section',
+  'https://cdn.example.com/photo.webp#fragment',
+] as const;
+
+const IMAGE_URLS_WITH_QUERY_AND_FRAGMENT = [
+  'https://cdn.example.com/photo.png?height=200#section',
+] as const;
+
+const STILL_VALID_PLAIN_URLS = [
+  'https://cdn.example.com/photo.png',
+  'https://cdn.example.com/photo.PNG',
+  'https://cdn.example.com/photo.jpeg',
+  'https://cdn.example.com/photo.jpg',
+  'https://cdn.example.com/photo.webp',
+  'https://cdn.example.com/photo.gif',
+  'data:image/png;base64,AAECAw==',
+] as const;
+
+const STILL_INVALID_URLS = [
+  'https://example.com/not-an-image.txt',
+  'ftp://example.com/photo.png',
+  'https://example.com/document.pdf',
+] as const;
+
+/**
+ * Additional defensive edge cases for the same failure mode.
+ *
+ * These cover real-world image URL shapes commonly seen with CDNs and
+ * pre-signed object storage URLs (S3, GCS, R2, etc.). They are deliberately
+ * tightly scoped to the same `?...` / `#...` suffix relaxation — they should
+ * NOT pass under any pattern that drops the extension requirement entirely.
+ */
+const DEFENSIVE_EXTRA_VALID_URLS = [
+  // URL-encoded query value (typical for signed URLs)
+  'https://cdn.example.com/photo.png?token=abc%20def%2F123',
+  // Pre-signed S3-style URL with multiple query params
+  'https://my-bucket.s3.amazonaws.com/photo.jpg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Expires=900&X-Amz-Signature=deadbeef',
+  // Both query and fragment, plus uppercase extension
+  'https://cdn.example.com/photo.PNG?x=1#frag',
+  // Empty query string (still a query, just no params)
+  'https://cdn.example.com/photo.gif?',
+  // Empty fragment
+  'https://cdn.example.com/photo.webp#',
+] as const;
+
+/**
+ * Defensive cases that must REMAIN rejected after the fix.
+ *
+ * The most important one is the "extension-in-the-path" shape:
+ * `https://cdn.example.com/some.png/redirect`. Naive relaxations of the
+ * regex (e.g. dropping the `$` anchor or replacing it with `.*`) would
+ * accept this, which would regress the explicit `image/*` allowlist.
+ */
+const DEFENSIVE_EXTRA_INVALID_URLS = [
+  // Extension is in the path, not terminal — must stay rejected
+  'https://cdn.example.com/some.png/redirect',
+  // Extension followed by another path segment
+  'https://cdn.example.com/photo.jpg/thumbnail',
+  // No extension at all, just a query string
+  'https://cdn.example.com/photo?type=png',
+  // Wrong scheme with query params
+  'ftp://example.com/photo.png?x=1',
+] as const;
+
+function firstImagePattern(supportedUrls: Record<string, RegExp[]>): RegExp[] {
+  const patterns = supportedUrls['image/*'];
+  if (!patterns) {
+    throw new Error('image/* patterns missing');
+  }
+  return patterns;
+}
+
+function matchesAny(url: string, patterns: RegExp[]): boolean {
+  return patterns.some((pattern) => pattern.test(url));
+}
+
+describe('Issue #484: image URL regex must accept query strings and fragments', () => {
+  describe('chat model supportedUrls', () => {
+    const patterns = firstImagePattern(chatModel.supportedUrls);
+    it.each(IMAGE_URLS_WITH_QUERY_PARAMS)('accepts query params: %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(true);
+    });
+    it.each(IMAGE_URLS_WITH_FRAGMENTS)('accepts fragment: %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(true);
+    });
+    it.each(
+      IMAGE_URLS_WITH_QUERY_AND_FRAGMENT,
+    )('accepts query+fragment: %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(true);
+    });
+    it.each(STILL_VALID_PLAIN_URLS)('still accepts plain: %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(true);
+    });
+    it.each(STILL_INVALID_URLS)('still rejects: %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(false);
+    });
+    it.each(DEFENSIVE_EXTRA_VALID_URLS)('defensive: accepts %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(true);
+    });
+    it.each(
+      DEFENSIVE_EXTRA_INVALID_URLS,
+    )('defensive: still rejects %s', (url) => {
+      expect(matchesAny(url, patterns)).toBe(false);
+    });
+  });
+
+  describe('completion model supportedUrls', () => {
+    const patterns = firstImagePattern(completionModel.supportedUrls);
+    it('accepts image URL with query params', () => {
+      expect(
+        matchesAny('https://cdn.example.com/photo.png?height=200', patterns),
+      ).toBe(true);
+    });
+    it('accepts image URL with fragment', () => {
+      expect(matchesAny('https://cdn.example.com/photo.webp#f', patterns)).toBe(
+        true,
+      );
+    });
+    it('still rejects non-image URL', () => {
+      expect(matchesAny('https://example.com/document.pdf', patterns)).toBe(
+        false,
+      );
+    });
+    it('defensive: still rejects extension-in-path', () => {
+      expect(
+        matchesAny('https://cdn.example.com/some.png/redirect', patterns),
+      ).toBe(false);
+    });
+    it('defensive: accepts pre-signed URL with multiple query params', () => {
+      expect(
+        matchesAny(
+          'https://my-bucket.s3.amazonaws.com/photo.jpg?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Expires=900&X-Amz-Signature=deadbeef',
+          patterns,
+        ),
+      ).toBe(true);
+    });
+  });
+});

src/chat/index.ts

@@ -70,7 +70,7 @@ export class OpenRouterChatLanguageModel implements LanguageModelV3 {
   readonly supportedUrls: Record<string, RegExp[]> = {
     'image/*': [
       /^data:image\/[a-zA-Z]+;base64,/,
-      /^https?:\/\/.+\.(jpg|jpeg|png|gif|webp)$/i,
+      /^https?:\/\/.+\.(jpg|jpeg|png|gif|webp)(?:[?#].*)?$/i,
     ],
     // 'text/*': [/^data:text\//, /^https?:\/\/.+$/],
     'application/*': [/^data:application\//, /^https?:\/\/.+$/],

src/completion/index.ts

@@ -54,7 +54,7 @@ export class OpenRouterCompletionLanguageModel implements LanguageModelV3 {
   readonly supportedUrls: Record<string, RegExp[]> = {
     'image/*': [
       /^data:image\/[a-zA-Z]+;base64,/,
-      /^https?:\/\/.+\.(jpg|jpeg|png|gif|webp)$/i,
+      /^https?:\/\/.+\.(jpg|jpeg|png|gif|webp)(?:[?#].*)?$/i,
     ],
     'text/*': [/^data:text\//, /^https?:\/\/.+$/],
     'application/*': [/^data:application\//, /^https?:\/\/.+$/],