@tuturuuu/ai 0.8.0 → 0.8.2

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@tuturuuu/ai",
   "license": "MIT",
-  "version": "0.8.0",
+  "version": "0.8.2",
   "repository": {
     "type": "git",
     "url": "https://github.com/tutur3u/platform",
@@ -22,34 +22,34 @@
     "type-check": "tsc --project tsconfig.typecheck.json"
   },
   "dependencies": {
-    "@ai-sdk/amazon-bedrock": "^5.0.42",
-    "@ai-sdk/anthropic": "^4.0.29",
-    "@ai-sdk/azure": "^4.0.30",
-    "@ai-sdk/cerebras": "^3.0.22",
-    "@ai-sdk/cohere": "^4.0.20",
-    "@ai-sdk/deepgram": "^3.0.20",
-    "@ai-sdk/deepinfra": "^3.0.22",
-    "@ai-sdk/deepseek": "^3.0.21",
-    "@ai-sdk/elevenlabs": "^3.0.21",
-    "@ai-sdk/fal": "^3.0.21",
-    "@ai-sdk/fireworks": "^3.0.24",
-    "@ai-sdk/gateway": "^4.0.40",
-    "@ai-sdk/gladia": "^3.0.20",
-    "@ai-sdk/google": "^4.0.33",
-    "@ai-sdk/google-vertex": "^5.0.40",
-    "@ai-sdk/groq": "^4.0.21",
-    "@ai-sdk/hume": "^3.0.20",
-    "@ai-sdk/lmnt": "^3.0.20",
-    "@ai-sdk/luma": "^3.0.21",
-    "@ai-sdk/mistral": "^4.0.22",
-    "@ai-sdk/openai": "^4.0.29",
-    "@ai-sdk/openai-compatible": "^3.0.22",
-    "@ai-sdk/perplexity": "^4.0.22",
-    "@ai-sdk/react": "^4.0.54",
-    "@ai-sdk/replicate": "^3.0.21",
-    "@ai-sdk/revai": "^3.0.20",
-    "@ai-sdk/togetherai": "^3.0.23",
-    "@ai-sdk/xai": "^4.0.27",
+    "@ai-sdk/amazon-bedrock": "^5.0.48",
+    "@ai-sdk/anthropic": "^4.0.34",
+    "@ai-sdk/azure": "^4.0.35",
+    "@ai-sdk/cerebras": "^3.0.25",
+    "@ai-sdk/cohere": "^4.0.23",
+    "@ai-sdk/deepgram": "^3.0.23",
+    "@ai-sdk/deepinfra": "^3.0.25",
+    "@ai-sdk/deepseek": "^3.0.24",
+    "@ai-sdk/elevenlabs": "^3.0.24",
+    "@ai-sdk/fal": "^3.0.24",
+    "@ai-sdk/fireworks": "^3.0.27",
+    "@ai-sdk/gateway": "^4.0.44",
+    "@ai-sdk/gladia": "^3.0.23",
+    "@ai-sdk/google": "^4.0.37",
+    "@ai-sdk/google-vertex": "^5.0.45",
+    "@ai-sdk/groq": "^4.0.24",
+    "@ai-sdk/hume": "^3.0.23",
+    "@ai-sdk/lmnt": "^3.0.23",
+    "@ai-sdk/luma": "^3.0.24",
+    "@ai-sdk/mistral": "^4.0.25",
+    "@ai-sdk/openai": "^4.0.34",
+    "@ai-sdk/openai-compatible": "^3.0.25",
+    "@ai-sdk/perplexity": "^4.0.25",
+    "@ai-sdk/react": "^4.0.59",
+    "@ai-sdk/replicate": "^3.0.24",
+    "@ai-sdk/revai": "^3.0.23",
+    "@ai-sdk/togetherai": "^3.0.26",
+    "@ai-sdk/xai": "^4.0.31",
     "@beeper/chat-adapter-matrix": "^0.2.0",
     "@bitbasti/chat-adapter-webex": "^0.1.0",
     "@chat-adapter/discord": "^4.36.0",
@@ -68,7 +68,7 @@
     "@chat-adapter/whatsapp": "^4.36.0",
     "@json-render/core": "^0.19.0",
     "@json-render/react": "^0.19.0",
-    "@liveblocks/chat-sdk-adapter": "^3.23.0",
+    "@liveblocks/chat-sdk-adapter": "^3.23.1",
     "@octokit/rest": "^22.0.1",
     "@resend/chat-sdk-adapter": "^0.2.2",
     "@streamdown/cjk": "^1.0.3",
@@ -76,12 +76,12 @@
     "@streamdown/math": "^1.0.2",
     "@streamdown/mermaid": "^1.0.2",
     "@tuturuuu/google": "0.1.0",
-    "@tuturuuu/internal-api": "0.28.0",
+    "@tuturuuu/internal-api": "0.29.1",
     "@tuturuuu/supabase": "0.5.0",
-    "@tuturuuu/utils": "0.22.0",
+    "@tuturuuu/utils": "0.24.2",
     "@vercel/sandbox": "^2.9.2",
     "@zernio/chat-sdk-adapter": "^0.5.0",
-    "ai": "^7.0.51",
+    "ai": "^7.0.56",
     "bash-tool": "^1.3.18",
     "chat": "^4.36.0",
     "chat-adapter-baileys": "^2.1.0",

package/src/chat/google/message-file-processing.ts

@@ -8,12 +8,119 @@ type ChatFile = {
 };
 
 const FILE_DOWNLOAD_CONCURRENCY = 4;
+// Supabase storage `list` defaults to 100 rows, so a chat that accumulated more
+// resources than this silently lost everything past the first page.
+const STORAGE_LIST_PAGE_SIZE = 100;
+const STORAGE_LIST_MAX_PAGES = 20;
+// Every request re-sends the chat's whole resource set, so an unbounded chat
+// would eventually build a request no provider will accept. Attachments are
+// admitted oldest-first until the budget is spent.
+const MAX_INJECTED_ATTACHMENT_BYTES = 32 * 1024 * 1024;
+
+// Storage only records a media type when the uploader knew one. Without this,
+// an image whose content type was lost on the way in reads as
+// `application/octet-stream`, misses the `image/` check, and reaches the model
+// as "this format cannot be passed directly" — invisible, with no error.
+const MEDIA_TYPE_BY_EXTENSION: Record<string, string> = {
+  aac: 'audio/aac',
+  avif: 'image/avif',
+  csv: 'text/csv',
+  flac: 'audio/flac',
+  gif: 'image/gif',
+  heic: 'image/heic',
+  jpeg: 'image/jpeg',
+  jpg: 'image/jpeg',
+  json: 'application/json',
+  m4a: 'audio/mp4',
+  md: 'text/markdown',
+  mov: 'video/quicktime',
+  mp3: 'audio/mpeg',
+  mp4: 'video/mp4',
+  ogg: 'audio/ogg',
+  opus: 'audio/ogg',
+  pdf: 'application/pdf',
+  png: 'image/png',
+  txt: 'text/plain',
+  wav: 'audio/wav',
+  webm: 'video/webm',
+  webp: 'image/webp',
+};
+
+const UNKNOWN_MEDIA_TYPE = 'application/octet-stream';
+
+export function resolveAttachmentMediaType(
+  fileName: string,
+  metadataMediaType?: string | null
+): string {
+  if (metadataMediaType && metadataMediaType !== UNKNOWN_MEDIA_TYPE) {
+    return metadataMediaType;
+  }
+
+  const extension = fileName.split('.').pop()?.toLowerCase();
+  if (extension && MEDIA_TYPE_BY_EXTENSION[extension]) {
+    return MEDIA_TYPE_BY_EXTENSION[extension];
+  }
+
+  return metadataMediaType || UNKNOWN_MEDIA_TYPE;
+}
 
 function maskIdentifier(value: string): string {
   if (value.length <= 8) return value;
   return `${value.slice(0, 4)}...${value.slice(-4)}`;
 }
 
+type StorageListClient = {
+  storage: {
+    from: (bucket: string) => {
+      list: (
+        path: string,
+        options: Record<string, unknown>
+      ) => Promise<{
+        data:
+          | { name: string; metadata?: Record<string, unknown> | null }[]
+          | null;
+        error: unknown;
+      }>;
+    };
+  };
+};
+
+export async function listAllChatResourceFiles(
+  sbAdmin: StorageListClient,
+  storagePath: string
+) {
+  const files: { name: string; metadata?: Record<string, unknown> | null }[] =
+    [];
+
+  for (let page = 0; page < STORAGE_LIST_MAX_PAGES; page += 1) {
+    const { data, error } = await sbAdmin.storage
+      .from('workspaces')
+      .list(storagePath, {
+        limit: STORAGE_LIST_PAGE_SIZE,
+        offset: page * STORAGE_LIST_PAGE_SIZE,
+        sortBy: { column: 'created_at', order: 'asc' },
+      });
+
+    if (error) {
+      console.error('Error listing files:', error);
+      // Keep whatever paged in rather than dropping the whole set.
+      return { files, truncated: false };
+    }
+
+    if (!data || data.length === 0) break;
+
+    files.push(...data);
+    if (data.length < STORAGE_LIST_PAGE_SIZE) {
+      return { files, truncated: false };
+    }
+  }
+
+  return {
+    files,
+    truncated: files.length >= STORAGE_LIST_PAGE_SIZE * STORAGE_LIST_MAX_PAGES,
+  };
+}
+
 async function getAllChatFiles(
   wsId: string,
   chatId: string,
@@ -23,24 +130,19 @@ async function getAllChatFiles(
     const sbAdmin = await createAdminClient();
 
     const storagePath = `${wsId}/chats/ai/resources/${chatId}`;
-    const { data: files, error: listError } = await sbAdmin.storage
-      .from('workspaces')
-      .list(storagePath, {
-        sortBy: { column: 'created_at', order: 'asc' },
-      });
-
-    if (listError) {
-      console.error('Error listing files:', listError);
-      return [];
-    }
+    const { files, truncated } = await listAllChatResourceFiles(
+      sbAdmin,
+      storagePath
+    );
 
     console.info('[Google Chat Files] listed chat files', {
       wsId: maskIdentifier(wsId),
       chatId: maskIdentifier(chatId),
-      fileCount: files?.length ?? 0,
+      fileCount: files.length,
+      truncated,
     });
 
-    if (!files || files.length === 0) {
+    if (files.length === 0) {
       return [];
     }
 
@@ -57,10 +159,11 @@ async function getAllChatFiles(
 
           const file = files[currentIndex]!;
           const fileName = file.name || 'unknown';
-          const mediaType =
-            file.metadata?.mediaType ||
-            file.metadata?.mimetype ||
-            'application/octet-stream';
+          const mediaType = resolveAttachmentMediaType(
+            fileName,
+            (file.metadata?.mediaType as string | undefined) ||
+              (file.metadata?.mimetype as string | undefined)
+          );
 
           const { data: fileData, error: downloadError } = await sbAdmin.storage
             .from('workspaces')
@@ -116,6 +219,14 @@ function addFilesToContent(
     'text/csv',
     'application/json',
     'text/markdown',
+    // Gemini accepts audio natively; without these an attached recording
+    // reached the model only as a "cannot be passed directly" note.
+    'audio/aac',
+    'audio/flac',
+    'audio/mp4',
+    'audio/mpeg',
+    'audio/ogg',
+    'audio/wav',
   ]);
 
   if (typeof existingContent === 'string') {
@@ -132,8 +243,21 @@ function addFilesToContent(
     }
   }
 
+  let injectedBytes = 0;
+  let skippedForBudget = 0;
+
   for (const file of chatFiles) {
     const { content, mediaType, fileName } = file;
+    const byteLength =
+      content instanceof ArrayBuffer ? content.byteLength : content.length;
+
+    // Oldest-first: an early attachment the conversation is actually about
+    // should not be evicted by a later one.
+    if (injectedBytes + byteLength > MAX_INJECTED_ATTACHMENT_BYTES) {
+      skippedForBudget += 1;
+      continue;
+    }
+    injectedBytes += byteLength;
 
     if (mediaType.startsWith('image/')) {
       const imagePart: ImagePart = {
@@ -172,6 +296,14 @@ function addFilesToContent(
     }
   }
 
+  if (skippedForBudget > 0) {
+    // Say so rather than answering as though nothing was attached.
+    contentParts.push({
+      type: 'text',
+      text: `${skippedForBudget} older attachment(s) were omitted from this request because the conversation exceeds the attachment size budget. Ask the user to re-attach anything you need.`,
+    });
+  }
+
   return contentParts;
 }
 

package/src/meetings/transcription/input.ts

@@ -0,0 +1,60 @@
+export const MAX_INLINE_TRANSCRIPTION_AUDIO_BYTES = 18 * 1024 * 1024;
+export const TRANSCRIPTION_MULTIPART_HEADROOM_BYTES = 1024 * 1024;
+export const MAX_TRANSCRIPTION_MULTIPART_REQUEST_BYTES =
+  MAX_INLINE_TRANSCRIPTION_AUDIO_BYTES + TRANSCRIPTION_MULTIPART_HEADROOM_BYTES;
+
+const TRANSCRIPTION_AUDIO_MEDIA_TYPES = [
+  'audio/webm',
+  'audio/mp4',
+  'audio/ogg',
+  'audio/mpeg',
+] as const;
+
+export type TranscriptionAudioMediaType =
+  (typeof TRANSCRIPTION_AUDIO_MEDIA_TYPES)[number];
+
+type TranscriptionAudioValidation =
+  | { mediaType: TranscriptionAudioMediaType; ok: true }
+  | { message: string; ok: false; status: 400 | 413 | 415 };
+
+export function validateTranscriptionAudioFile(
+  audio: Pick<File, 'size' | 'type'>
+): TranscriptionAudioValidation {
+  if (audio.size <= 0) {
+    return { message: 'The recording is empty.', ok: false, status: 400 };
+  }
+
+  if (audio.size > MAX_INLINE_TRANSCRIPTION_AUDIO_BYTES) {
+    return {
+      message: 'The recording exceeds the inline transcription limit.',
+      ok: false,
+      status: 413,
+    };
+  }
+
+  const baseMediaType = audio.type.split(';', 1)[0]?.trim().toLowerCase();
+  const mediaType = TRANSCRIPTION_AUDIO_MEDIA_TYPES.find(
+    (allowedMediaType) => allowedMediaType === baseMediaType
+  );
+  if (!mediaType) {
+    return {
+      message:
+        'The recording format is not supported for inline transcription.',
+      ok: false,
+      status: 415,
+    };
+  }
+
+  return { mediaType, ok: true };
+}
+
+export function hasOversizedDeclaredContentLength(req: Request) {
+  const contentLength = req.headers.get('content-length')?.trim();
+  if (!contentLength || !/^\d+$/u.test(contentLength)) {
+    return false;
+  }
+
+  return (
+    BigInt(contentLength) > BigInt(MAX_TRANSCRIPTION_MULTIPART_REQUEST_BYTES)
+  );
+}

package/src/meetings/transcription/route.ts

@@ -3,6 +3,10 @@ import { createClient } from '@tuturuuu/supabase/next/server';
 import { generateObject } from 'ai';
 import { z } from 'zod';
 import { withAiMemory } from '../../memory';
+import {
+  hasOversizedDeclaredContentLength,
+  validateTranscriptionAudioFile,
+} from './input';
 
 const DEFAULT_MODEL_NAME = 'gemini-3.1-flash-lite';
 
@@ -42,14 +46,23 @@ export function createPOST() {
         return new Response('Unauthorized', { status: 401 });
       }
 
+      if (hasOversizedDeclaredContentLength(req)) {
+        return new Response('Audio file too large', { status: 413 });
+      }
+
       const formData = await req.formData();
-      const audioFile = formData.get('audio') as File;
+      const audioFile = formData.get('audio');
       const wsId = formData.get('wsId') as string | null;
 
-      if (!audioFile) {
+      if (!(audioFile instanceof File)) {
         return new Response('No audio file provided', { status: 400 });
       }
 
+      const validation = validateTranscriptionAudioFile(audioFile);
+      if (!validation.ok) {
+        return new Response(validation.message, { status: validation.status });
+      }
+
       // Convert file to buffer
       const audioBuffer = await audioFile.arrayBuffer();
       const audioUint8Array = new Uint8Array(audioBuffer);
@@ -73,7 +86,7 @@ export function createPOST() {
             content: [
               {
                 type: 'file',
-                mediaType: 'audio/mpeg',
+                mediaType: validation.mediaType,
                 data: audioUint8Array,
               },
             ],