📑 feat: Anthropic Direct Provider Upload (#9072)

* feat: implement Anthropic native PDF support with document preservation

- Add comprehensive debug logging throughout PDF processing pipeline
- Refactor attachment processing to separate image and document handling
- Create distinct addImageURLs(), addDocuments(), and processAttachments() methods
- Fix critical bugs in stream handling and parameter passing
- Add streamToBuffer utility for proper stream-to-buffer conversion
- Remove api/agents submodule from repository

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <noreply@anthropic.com>

* chore: remove out of scope formatting changes

* fix: stop duplication of file in chat on end of response stream

* chore: bring back file search and ocr options

* chore: localize upload to provider string in file menu

* refactor: change createMenuItems args to fit new pattern introduced by anthropic-native-pdf-support

* feat: add cache point for pdfs processed by anthropic endpoint since they are unlikely to change and should benefit from caching

* feat: combine Upload Image into Upload to Provider since they both perform direct upload and change provider upload icon to reflect multimodal upload

* feat: add citations support according to docs

* refactor: remove redundant 'document' check since documents are handled properly by formatMessage in the agents repo now

* refactor: change upload logic so anthropic endpoint isn't exempted from normal upload path using Agents for consistency with the rest of the upload logic

* fix: include width and height in return from uploadLocalFile so images are correctly identified when going through an AgentUpload in addImageURLs

* chore: remove client specific handling since the direct provider stuff is handled by the agent client

* feat: handle documents in AgentClient so no need for change to agents repo

* chore: removed unused changes

* chore: remove auto generated comments from OG commit

* feat: add logic for agents to use direct to provider uploads if supported (currently just anthropic)

* fix: reintroduce role check to fix render error because of undefined value for Content Part

* fix: actually fix render bug by using proper isCreatedByUser check and making sure our mutation of formattedMessage.content is consistent

---------

Co-authored-by: Andres Restrepo <andres@thelinuxkid.com>
Co-authored-by: Claude <noreply@anthropic.com>

api/server/controllers/agents/client.js

@@ -222,6 +222,42 @@ class AgentClient extends BaseClient {
     return files;
   }
 
+  async addDocuments(message, attachments) {
+    const documentResult =
+      await require('~/server/services/Files/documents').encodeAndFormatDocuments(
+        this.options.req,
+        attachments,
+        this.options.agent.provider,
+      );
+    message.documents =
+      documentResult.documents && documentResult.documents.length
+        ? documentResult.documents
+        : undefined;
+    return documentResult.files;
+  }
+
+  async processAttachments(message, attachments) {
+    const [imageFiles, documentFiles] = await Promise.all([
+      this.addImageURLs(message, attachments),
+      this.addDocuments(message, attachments),
+    ]);
+
+    const allFiles = [...imageFiles, ...documentFiles];
+    const seenFileIds = new Set();
+    const uniqueFiles = [];
+
+    for (const file of allFiles) {
+      if (file.file_id && !seenFileIds.has(file.file_id)) {
+        seenFileIds.add(file.file_id);
+        uniqueFiles.push(file);
+      } else if (!file.file_id) {
+        uniqueFiles.push(file);
+      }
+    }
+
+    return uniqueFiles;
+  }
+
   async buildMessages(
     messages,
     parentMessageId,
@@ -255,7 +291,7 @@ class AgentClient extends BaseClient {
         };
       }
 
-      const files = await this.addImageURLs(
+      const files = await this.processAttachments(
         orderedMessages[orderedMessages.length - 1],
         attachments,
       );
@@ -278,6 +314,27 @@ class AgentClient extends BaseClient {
         assistantName: this.options?.modelLabel,
       });
 
+      if (
+        message.documents &&
+        message.documents.length > 0 &&
+        message.isCreatedByUser &&
+        this.options.agent.provider === EModelEndpoint.anthropic
+      ) {
+        const contentParts = [];
+        contentParts.push(...message.documents);
+        if (message.image_urls && message.image_urls.length > 0) {
+          contentParts.push(...message.image_urls);
+        }
+
+        if (typeof formattedMessage.content === 'string') {
+          contentParts.push({ type: 'text', text: formattedMessage.content });
+        } else {
+          const textPart = formattedMessage.content.find((part) => part.type === 'text');
+          contentParts.push(textPart);
+        }
+        formattedMessage.content = contentParts;
+      }
+
       if (message.ocr && i !== orderedMessages.length - 1) {
         if (typeof formattedMessage.content === 'string') {
           formattedMessage.content = message.ocr + '\n' + formattedMessage.content;
@@ -793,6 +850,7 @@ class AgentClient extends BaseClient {
       };
 
       const toolSet = new Set((this.options.agent.tools ?? []).map((tool) => tool && tool.name));
+
       let { messages: initialMessages, indexTokenCountMap } = formatAgentMessages(
         payload,
         this.indexTokenCountMap,

api/server/services/Files/Local/crud.js

@@ -4,6 +4,7 @@ const axios = require('axios');
 const { logger } = require('@librechat/data-schemas');
 const { EModelEndpoint } = require('librechat-data-provider');
 const { generateShortLivedToken } = require('@librechat/api');
+const { resizeImageBuffer } = require('~/server/services/Files/images/resize');
 const { getBufferMetadata } = require('~/server/utils');
 const paths = require('~/config/paths');
 
@@ -286,7 +287,18 @@ async function uploadLocalFile({ req, file, file_id }) {
   await fs.promises.writeFile(newPath, inputBuffer);
   const filepath = path.posix.join('/', 'uploads', req.user.id, path.basename(newPath));
 
-  return { filepath, bytes };
+  let height, width;
+  if (file.mimetype && file.mimetype.startsWith('image/')) {
+    try {
+      const { width: imgWidth, height: imgHeight } = await resizeImageBuffer(inputBuffer, 'high');
+      height = imgHeight;
+      width = imgWidth;
+    } catch (error) {
+      logger.warn('[uploadLocalFile] Could not get image dimensions:', error.message);
+    }
+  }
+
+  return { filepath, bytes, height, width };
 }
 
 /**

api/server/services/Files/documents/encode.js

@@ -0,0 +1,164 @@
+const { EModelEndpoint } = require('librechat-data-provider');
+const { getStrategyFunctions } = require('~/server/services/Files/strategies');
+const { validateAnthropicPdf } = require('../validation/pdfValidator');
+
+/**
+ * Converts a readable stream to a buffer.
+ *
+ * @param {NodeJS.ReadableStream} stream - The readable stream to convert.
+ * @returns {Promise<Buffer>} - Promise resolving to the buffer.
+ */
+async function streamToBuffer(stream) {
+  return new Promise((resolve, reject) => {
+    const chunks = [];
+
+    stream.on('data', (chunk) => {
+      chunks.push(chunk);
+    });
+
+    stream.on('end', () => {
+      try {
+        const buffer = Buffer.concat(chunks);
+        chunks.length = 0;
+        resolve(buffer);
+      } catch (err) {
+        reject(err);
+      }
+    });
+
+    stream.on('error', (error) => {
+      chunks.length = 0;
+      reject(error);
+    });
+  }).finally(() => {
+    if (stream.destroy && typeof stream.destroy === 'function') {
+      stream.destroy();
+    }
+  });
+}
+
+/**
+ * Processes and encodes document files for various endpoints
+ *
+ * @param {Express.Request} req - Express request object
+ * @param {MongoFile[]} files - Array of file objects to process
+ * @param {string} endpoint - The endpoint identifier (e.g., EModelEndpoint.anthropic)
+ * @returns {Promise<{documents: MessageContentDocument[], files: MongoFile[]}>}
+ */
+async function encodeAndFormatDocuments(req, files, endpoint) {
+  const promises = [];
+  /** @type {Record<FileSources, Pick<ReturnType<typeof getStrategyFunctions>, 'prepareDocumentPayload' | 'getDownloadStream'>>} */
+  const encodingMethods = {};
+  /** @type {{ documents: MessageContentDocument[]; files: MongoFile[] }} */
+  const result = {
+    documents: [],
+    files: [],
+  };
+
+  if (!files || !files.length) {
+    return result;
+  }
+
+  const documentFiles = files.filter(
+    (file) => file.type === 'application/pdf' || file.type?.startsWith('application/'), // Future: support for other document types
+  );
+
+  if (!documentFiles.length) {
+    return result;
+  }
+
+  for (let file of documentFiles) {
+    /** @type {FileSources} */
+    const source = file.source ?? 'local';
+
+    if (file.type !== 'application/pdf' || endpoint !== EModelEndpoint.anthropic) {
+      continue;
+    }
+
+    if (!encodingMethods[source]) {
+      encodingMethods[source] = getStrategyFunctions(source);
+    }
+
+    const fileMetadata = {
+      file_id: file.file_id || file._id,
+      temp_file_id: file.temp_file_id,
+      filepath: file.filepath,
+      source: file.source,
+      filename: file.filename,
+      type: file.type,
+    };
+
+    promises.push([file, fileMetadata]);
+  }
+
+  const results = await Promise.allSettled(
+    promises.map(async ([file, fileMetadata]) => {
+      if (!file || !fileMetadata) {
+        return { file: null, content: null, metadata: fileMetadata };
+      }
+
+      try {
+        const source = file.source ?? 'local';
+        const { getDownloadStream } = encodingMethods[source];
+
+        const stream = await getDownloadStream(req, file.filepath);
+        const buffer = await streamToBuffer(stream);
+        const documentContent = buffer.toString('base64');
+
+        return {
+          file,
+          content: documentContent,
+          metadata: fileMetadata,
+        };
+      } catch (error) {
+        console.error(`Error processing document ${file.filename}:`, error);
+        return { file, content: null, metadata: fileMetadata };
+      }
+    }),
+  );
+
+  for (const settledResult of results) {
+    if (settledResult.status === 'rejected') {
+      console.error('Document processing failed:', settledResult.reason);
+      continue;
+    }
+
+    const { file, content, metadata } = settledResult.value;
+
+    if (!content || !file) {
+      if (metadata) {
+        result.files.push(metadata);
+      }
+      continue;
+    }
+
+    if (file.type === 'application/pdf' && endpoint === EModelEndpoint.anthropic) {
+      const pdfBuffer = Buffer.from(content, 'base64');
+      const validation = await validateAnthropicPdf(pdfBuffer, pdfBuffer.length);
+
+      if (!validation.isValid) {
+        throw new Error(`PDF validation failed: ${validation.error}`);
+      }
+
+      const documentPart = {
+        type: 'document',
+        source: {
+          type: 'base64',
+          media_type: 'application/pdf',
+          data: content,
+        },
+        cache_control: { type: 'ephemeral' },
+        citations: { enabled: true },
+      };
+
+      result.documents.push(documentPart);
+      result.files.push(metadata);
+    }
+  }
+
+  return result;
+}
+
+module.exports = {
+  encodeAndFormatDocuments,
+};

api/server/services/Files/documents/index.js

@@ -0,0 +1,5 @@
+const { encodeAndFormatDocuments } = require('./encode');
+
+module.exports = {
+  encodeAndFormatDocuments,
+};

api/server/services/Files/process.js

@@ -419,11 +419,11 @@ const processFileUpload = async ({ req, res, metadata }) => {
   const isAssistantUpload = isAssistantsEndpoint(metadata.endpoint);
   const assistantSource =
     metadata.endpoint === EModelEndpoint.azureAssistants ? FileSources.azure : FileSources.openai;
+
   // Use the configured file strategy for regular file uploads (not vectordb)
   const source = isAssistantUpload ? assistantSource : appConfig.fileStrategy;
   const { handleFileUpload } = getStrategyFunctions(source);
   const { file_id, temp_file_id = null } = metadata;
-
   /** @type {OpenAI | undefined} */
   let openai;
   if (checkOpenAIStorage(source)) {

api/server/services/Files/validation/pdfValidator.js

@@ -0,0 +1,72 @@
+const { logger } = require('~/config');
+const { anthropicPdfSizeLimit } = require('librechat-data-provider');
+
+/**
+ * Validates if a PDF meets Anthropic's requirements
+ * @param {Buffer} pdfBuffer - The PDF file as a buffer
+ * @param {number} fileSize - The file size in bytes
+ * @returns {Promise<{isValid: boolean, error?: string}>}
+ */
+async function validateAnthropicPdf(pdfBuffer, fileSize) {
+  try {
+    if (fileSize > anthropicPdfSizeLimit) {
+      return {
+        isValid: false,
+        error: `PDF file size (${Math.round(fileSize / (1024 * 1024))}MB) exceeds Anthropic's 32MB limit`,
+      };
+    }
+
+    if (!pdfBuffer || pdfBuffer.length < 5) {
+      return {
+        isValid: false,
+        error: 'Invalid PDF file: too small or corrupted',
+      };
+    }
+
+    const pdfHeader = pdfBuffer.subarray(0, 5).toString();
+    if (!pdfHeader.startsWith('%PDF-')) {
+      return {
+        isValid: false,
+        error: 'Invalid PDF file: missing PDF header',
+      };
+    }
+
+    const pdfContent = pdfBuffer.toString('binary');
+    if (
+      pdfContent.includes('/Encrypt ') ||
+      pdfContent.includes('/U (') ||
+      pdfContent.includes('/O (')
+    ) {
+      return {
+        isValid: false,
+        error: 'PDF is password-protected or encrypted. Anthropic requires unencrypted PDFs.',
+      };
+    }
+
+    const pageMatches = pdfContent.match(/\/Type[\s]*\/Page[^s]/g);
+    const estimatedPages = pageMatches ? pageMatches.length : 1;
+
+    if (estimatedPages > 100) {
+      return {
+        isValid: false,
+        error: `PDF has approximately ${estimatedPages} pages, exceeding Anthropic's 100-page limit`,
+      };
+    }
+
+    logger.debug(
+      `PDF validation passed: ${Math.round(fileSize / 1024)}KB, ~${estimatedPages} pages`,
+    );
+
+    return { isValid: true };
+  } catch (error) {
+    logger.error('PDF validation error:', error);
+    return {
+      isValid: false,
+      error: 'Failed to validate PDF file',
+    };
+  }
+}
+
+module.exports = {
+  validateAnthropicPdf,
+};

client/src/common/agents-types.ts

@@ -21,6 +21,7 @@ export type TAgentCapabilities = {
   [AgentCapabilities.execute_code]: boolean;
   [AgentCapabilities.end_after_tools]?: boolean;
   [AgentCapabilities.hide_sequential_outputs]?: boolean;
+  [AgentCapabilities.direct_upload]?: boolean;
 };
 
 export type AgentForm = {

client/src/components/Chat/Input/Files/AttachFileChat.tsx

@@ -36,6 +36,7 @@ function AttachFileChat({ disableInputs }: { disableInputs: boolean }) {
         disabled={disableInputs}
         conversationId={conversationId}
         endpointFileConfig={endpointFileConfig}
+        endpoint={endpoint}
       />
     );
   }

client/src/components/Chat/Input/Files/AttachFileMenu.tsx

@@ -1,7 +1,13 @@
 import React, { useRef, useState, useMemo } from 'react';
 import * as Ariakit from '@ariakit/react';
 import { useSetRecoilState } from 'recoil';
-import { FileSearch, ImageUpIcon, TerminalSquareIcon, FileType2Icon } from 'lucide-react';
+import {
+  FileSearch,
+  ImageUpIcon,
+  TerminalSquareIcon,
+  FileType2Icon,
+  FileImageIcon,
+} from 'lucide-react';
 import { EToolResources, EModelEndpoint, defaultAgentCapabilities } from 'librechat-data-provider';
 import {
   FileUpload,
@@ -14,8 +20,9 @@ import type { EndpointFileConfig } from 'librechat-data-provider';
 import { useLocalize, useGetAgentsConfig, useFileHandling, useAgentCapabilities } from '~/hooks';
 import useSharePointFileHandling from '~/hooks/Files/useSharePointFileHandling';
 import { SharePointPickerDialog } from '~/components/SharePoint';
-import { useGetStartupConfig } from '~/data-provider';
+import { useGetStartupConfig, useGetAgentByIdQuery } from '~/data-provider';
 import { ephemeralAgentByConvoId } from '~/store';
+import { useChatContext } from '~/Providers/ChatContext';
 import { MenuItemProps } from '~/common';
 import { cn } from '~/utils';
 
@@ -23,9 +30,15 @@ interface AttachFileMenuProps {
   conversationId: string;
   disabled?: boolean | null;
   endpointFileConfig?: EndpointFileConfig;
+  endpoint?: string | null;
 }
 
-const AttachFileMenu = ({ disabled, conversationId, endpointFileConfig }: AttachFileMenuProps) => {
+const AttachFileMenu = ({
+  disabled,
+  conversationId,
+  endpointFileConfig,
+  endpoint,
+}: AttachFileMenuProps) => {
   const localize = useLocalize();
   const isUploadDisabled = disabled ?? false;
   const inputRef = useRef<HTMLInputElement>(null);
@@ -46,34 +59,68 @@ const AttachFileMenu = ({ disabled, conversationId, endpointFileConfig }: Attach
 
   const [isSharePointDialogOpen, setIsSharePointDialogOpen] = useState(false);
   const { agentsConfig } = useGetAgentsConfig();
+  const { conversation } = useChatContext();
+
+  // Get agent details if using an agent
+  const { data: agent } = useGetAgentByIdQuery(conversation?.agent_id ?? '', {
+    enabled: !!conversation?.agent_id && conversation?.agent_id !== 'ephemeral',
+  });
+
   /** TODO: Ephemeral Agent Capabilities
    * Allow defining agent capabilities on a per-endpoint basis
    * Use definition for agents endpoint for ephemeral agents
    * */
   const capabilities = useAgentCapabilities(agentsConfig?.capabilities ?? defaultAgentCapabilities);
 
-  const handleUploadClick = (isImage?: boolean) => {
+  const handleUploadClick = (fileType?: 'image' | 'document' | 'anthropic_multimodal') => {
     if (!inputRef.current) {
       return;
     }
     inputRef.current.value = '';
-    inputRef.current.accept = isImage === true ? 'image/*' : '';
+    if (fileType === 'image') {
+      inputRef.current.accept = 'image/*';
+    } else if (fileType === 'document') {
+      inputRef.current.accept = '.pdf,application/pdf';
+    } else if (fileType === 'anthropic_multimodal') {
+      inputRef.current.accept = 'image/*,.pdf,application/pdf';
+    } else {
+      inputRef.current.accept = '';
+    }
     inputRef.current.click();
     inputRef.current.accept = '';
   };
 
   const dropdownItems = useMemo(() => {
-    const createMenuItems = (onAction: (isImage?: boolean) => void) => {
+    const createMenuItems = (
-      const items: MenuItemProps[] = [
+      onAction: (fileType?: 'image' | 'document' | 'anthropic_multimodal') => void,
-        {
+    ) => {
+      const items: MenuItemProps[] = [];
+
+      // this is temporary until i add direct upload support for the other providers and can make a more robust solution
+      const isAnthropicAgent = agent?.provider === 'anthropic';
+      const shouldShowDirectUpload = endpoint === EModelEndpoint.anthropic || isAnthropicAgent;
+
+      if (!shouldShowDirectUpload) {
+        items.push({
           label: localize('com_ui_upload_image_input'),
           onClick: () => {
             setToolResource(undefined);
-            onAction(true);
+            onAction('image');
           },
           icon: <ImageUpIcon className="icon-md" />,
-        },
+        });
-      ];
+      }
+
+      if (shouldShowDirectUpload) {
+        items.push({
+          label: localize('com_ui_upload_provider'),
+          onClick: () => {
+            setToolResource(EToolResources.direct_upload);
+            onAction('anthropic_multimodal');
+          },
+          icon: <FileImageIcon className="icon-md" />,
+        });
+      }
 
       if (capabilities.ocrEnabled) {
         items.push({
@@ -139,6 +186,7 @@ const AttachFileMenu = ({ disabled, conversationId, endpointFileConfig }: Attach
     setEphemeralAgent,
     sharePointEnabled,
     setIsSharePointDialogOpen,
+    endpoint,
   ]);
 
   const menuTrigger = (

client/src/hooks/Agents/useAgentCapabilities.ts

@@ -9,6 +9,7 @@ interface AgentCapabilitiesResult {
   fileSearchEnabled: boolean;
   webSearchEnabled: boolean;
   codeEnabled: boolean;
+  directUploadEnabled: boolean;
 }
 
 export default function useAgentCapabilities(
@@ -49,6 +50,11 @@ export default function useAgentCapabilities(
     [capabilities],
   );
 
+  const directUploadEnabled = useMemo(
+    () => capabilities?.includes(AgentCapabilities.direct_upload) ?? false,
+    [capabilities],
+  );
+
   return {
     ocrEnabled,
     codeEnabled,
@@ -57,5 +63,6 @@ export default function useAgentCapabilities(
     artifactsEnabled,
     webSearchEnabled,
     fileSearchEnabled,
+    directUploadEnabled,
   };
 }

client/src/locales/en/translation.json

@@ -1218,6 +1218,7 @@
   "com_ui_upload_invalid": "Invalid file for upload. Must be an image not exceeding the limit",
   "com_ui_upload_invalid_var": "Invalid file for upload. Must be an image not exceeding {{0}} MB",
   "com_ui_upload_ocr_text": "Upload as Text",
+  "com_ui_upload_provider": "Upload to Provider",
   "com_ui_upload_success": "Successfully uploaded file",
   "com_ui_upload_type": "Select Upload Type",
   "com_ui_usage": "Usage",

packages/data-provider/src/config.ts

@@ -175,6 +175,7 @@ export enum Capabilities {
 export enum AgentCapabilities {
   hide_sequential_outputs = 'hide_sequential_outputs',
   end_after_tools = 'end_after_tools',
+  direct_upload = 'direct_upload',
   execute_code = 'execute_code',
   file_search = 'file_search',
   web_search = 'web_search',
@@ -256,6 +257,7 @@ export const defaultAgentCapabilities = [
   AgentCapabilities.tools,
   AgentCapabilities.chain,
   AgentCapabilities.ocr,
+  AgentCapabilities.direct_upload,
 ];
 
 export const agentsEndpointSchema = baseEndpointSchema

packages/data-provider/src/file-config.ts

@@ -186,6 +186,10 @@ export const mbToBytes = (mb: number): number => mb * megabyte;
 
 const defaultSizeLimit = mbToBytes(512);
 const defaultTokenLimit = 100000;
+
+// Anthropic PDF limits: 32MB max, 100 pages max
+export const anthropicPdfSizeLimit = mbToBytes(32);
+
 const assistantsFileConfig = {
   fileLimit: 10,
   fileSizeLimit: defaultSizeLimit,
@@ -199,6 +203,14 @@ export const fileConfig = {
     [EModelEndpoint.assistants]: assistantsFileConfig,
     [EModelEndpoint.azureAssistants]: assistantsFileConfig,
     [EModelEndpoint.agents]: assistantsFileConfig,
+    [EModelEndpoint.anthropic]: {
+      fileLimit: 10,
+      fileSizeLimit: defaultSizeLimit,
+      totalSizeLimit: defaultSizeLimit,
+      supportedMimeTypes,
+      disabled: false,
+      pdfSizeLimit: anthropicPdfSizeLimit,
+    },
     default: {
       fileLimit: 10,
       fileSizeLimit: defaultSizeLimit,

packages/data-provider/src/types/assistants.ts

@@ -27,6 +27,7 @@ export enum Tools {
 
 export enum EToolResources {
   code_interpreter = 'code_interpreter',
+  direct_upload = 'direct_upload',
   execute_code = 'execute_code',
   file_search = 'file_search',
   image_edit = 'image_edit',