Fix Safari LiteRT chat: undefined is not a function

ai-chat.js

@@ -315,6 +315,55 @@ function extractLiteRTText(response) {
     .join('');
 }
 
+/**
+ * Consume LiteRT `sendMessageStreaming` output across browsers.
+ * Official API returns a `ReadableStream`. Chromium often supports
+ * `for await...of` on streams; Safari/WebKit frequently does not
+ * (`ReadableStream.prototype[Symbol.asyncIterator]` missing), which throws:
+ * `undefined is not a function (near '...s of a...')`.
+ * Prefer async iteration when present; otherwise use `getReader()`.
+ * @param {AsyncIterable|ReadableStream|Promise<AsyncIterable|ReadableStream>|null|undefined} streamLike
+ */
+export async function* iterateMessageStream(streamLike) {
+  let stream = streamLike;
+  if (stream != null && typeof stream.then === 'function') {
+    stream = await stream;
+  }
+  if (stream == null) return;
+
+  if (typeof stream[Symbol.asyncIterator] === 'function') {
+    yield* stream;
+    return;
+  }
+
+  if (typeof stream.getReader === 'function') {
+    const reader = stream.getReader();
+    try {
+      while (true) {
+        const { done, value } = await reader.read();
+        if (done) break;
+        yield value;
+      }
+    } finally {
+      try {
+        reader.releaseLock();
+      } catch {
+        // ignore — lock may already be released after close/error
+      }
+    }
+    return;
+  }
+
+  if (typeof stream[Symbol.iterator] === 'function') {
+    yield* stream;
+    return;
+  }
+
+  throw new TypeError(
+    'LiteRT streaming response is not an async iterable or ReadableStream',
+  );
+}
+
 /** Strip accidental thinking / turn markers from streamed text (defense in depth). */
 export function sanitizeModelReply(text) {
   if (!text) return '';
@@ -1022,8 +1071,11 @@ export class AiChat {
     let reply = '';
 
     if (typeof conversation.sendMessageStreaming === 'function') {
-      const stream = conversation.sendMessageStreaming(userText);
-      for await (const chunk of stream) {
+      // Do not `for await` the raw return value — it is a ReadableStream, and
+      // Safari cannot async-iterate ReadableStream (see iterateMessageStream).
+      for await (const chunk of iterateMessageStream(
+        conversation.sendMessageStreaming(userText),
+      )) {
         const delta = extractLiteRTText(chunk);
         if (!delta) continue;
         // Streaming chunks may be cumulative or incremental — prefer append of delta text pieces.

tests/unit/guardrails.spec.ts

@@ -252,6 +252,58 @@ test.describe('Guardrails Unit', () => {
     expect(result.createArgs.preface.messages[0].content).toContain('vd3-cbun');
   });
 
+  test('AiChat LiteRT generate works when stream is ReadableStream without asyncIterator (Safari)', async ({ page }) => {
+    const result = await page.evaluate(async () => {
+      const mod = await import('/ai-chat.js');
+
+      // Reader-only stream — no Symbol.asyncIterator (Safari/WebKit shape).
+      const readerOnlyStream = (chunks) => {
+        let i = 0;
+        return {
+          getReader() {
+            return {
+              async read() {
+                if (i >= chunks.length) return { done: true, value: undefined };
+                return { done: false, value: chunks[i++] };
+              },
+              releaseLock() {},
+            };
+          },
+        };
+      };
+
+      const collected = [];
+      for await (const chunk of mod.iterateMessageStream(
+        readerOnlyStream([
+          { content: [{ type: 'text', text: 'yo' }] },
+          { content: [{ type: 'text', text: ' man' }] },
+        ]),
+      )) {
+        collected.push(chunk);
+      }
+
+      const chat = new mod.AiChat({ modelId: 'gemma-4-E2B-it-web' });
+      chat._isLoaded = true;
+      chat.engine = {
+        createConversation: async () => ({
+          sendMessageStreaming: () => readerOnlyStream([
+            { content: [{ type: 'text', text: 'Hey' }] },
+            { content: [{ type: 'text', text: ' there' }] },
+          ]),
+          delete: async () => {},
+        }),
+      };
+      chat._conversation = await chat.engine.createConversation();
+      const updates = [];
+      const reply = await chat.generate('yo man', (t) => updates.push(t));
+      return { collectedLen: collected.length, reply, updates };
+    });
+
+    expect(result.collectedLen).toBe(2);
+    expect(result.reply).toBe('Hey there');
+    expect(result.updates.at(-1)).toBe('Hey there');
+  });
+
   test('AiChat Gemma 4 MLC payloads omit system role (WebLLM template limitation)', async ({ page }) => {
     const result = await page.evaluate(async () => {
       const mod = await import('/ai-chat.js');