@things-factory/ai-assistant 10.1.226 → 10.1.227

package/dist-client/components/assistant-chat.js

@@ -34,12 +34,14 @@ import { i18next } from '@operato/i18n';
 import { ScrollbarStyles } from '@operato/styles';
 import gql from 'graphql-tag';
 import { renderMarkdown, injectMentionRefids, injectMentionUserIds, stripMentionRefids } from './markdown.js';
-import { acceptOf, chatAttachmentUrl, refuseAttachments, textCharsOf } from '../utils/attachment-policy.js';
+import { acceptOf, chatAttachmentUrl, limitsOf, refuseAttachments, textCharsOf } from '../utils/attachment-policy.js';
 import { fileUploadLimits } from '@operato/utils/file-acceptance.js';
 import '@operato/input/ox-input-file.js';
 import { pastedTableFile } from '../utils/pasted-table.js';
 import { appendWords, composerHeight, composerKeyAction, composerSuggestion, sendShortcut, tableAsWords } from '../utils/composer-frame.js';
 import { readSpreadsheet, spreadsheetKindOf } from '../utils/spreadsheet-text.js';
+import { documentKindOf, readDocx, readPdf, readPptx } from '../utils/document-text.js';
+import { drawPdfPage, loadPdfjs } from '../utils/pdf-pages.js';
 import { entryReviewAddress, offersWorkspaceReview } from '../utils/entry-workspace.js';
 import { buildAssistantChatInput, buildChatMutationInput } from './chat-input-builder.js';
 import { agentDetail, offersAgentPicker, pickAgentId, rememberAgent, rememberedAgent } from '../utils/agent-choice.js';
@@ -514,6 +516,15 @@ let OxAssistantChat = class OxAssistantChat extends LitElement {
             }
             list = read.files;
         }
+        if (this.usesAttachmentPolicy && this.modeView?.attachments?.kinds?.includes('document')) {
+            /* A PDF, .docx or .pptx travels as its text with its page and slide numbers, a scanned page as an image (§9-6). */
+            const read = await this.readDocuments(list);
+            if ('refusal' in read) {
+                this.errorMessage = String(i18next.t(`ai-assistant.text.${read.refusal.code}`, { ...read.refusal, defaultValue: read.refusal.code }));
+                return;
+            }
+            list = read.files;
+        }
         if (this.usesAttachmentPolicy) {
             /* Said before anything is sent - never cut silently (§9-1). The one picking rule, the server's upload limit included. */
             const refusal = refuseAttachments(this.modeView?.attachments, [...this.attachments, ...list].map(f => ({ name: f.name, type: f.type, size: f.size })), await fileUploadLimits());
@@ -570,6 +581,41 @@ let OxAssistantChat = class OxAssistantChat extends LitElement {
         }
         return { files: out };
     }
+    /** Each document replaced by its text, and a PDF's scanned pages by their images; any other file passes as it is. */
+    async readDocuments(list) {
+        const policy = this.modeView?.attachments;
+        const maxChars = textCharsOf(policy);
+        const out = [];
+        for (const file of list) {
+            const kind = documentKindOf(file);
+            if (!kind) {
+                out.push(file);
+                continue;
+            }
+            if (kind === 'doc' || kind === 'ppt')
+                return { refusal: { code: 'attachment-old-office-not-read', name: file.name } };
+            const data = await file.arrayBuffer();
+            let read;
+            if (kind === 'pdf') {
+                const pdfjs = await loadPdfjs();
+                let pdf;
+                read = await readPdf(async () => (pdf = await pdfjs.getDocument({ data }).promise), file.name, maxChars, 
+                /* The text travels as one file; the scans may take the rest of what one message carries. */
+                { allowed: !!policy?.kinds?.includes('image'), max: Math.max(0, limitsOf(policy).maxFiles - 1) }, number => drawPdfPage(pdf, number));
+            }
+            else {
+                const mod = await import('jszip');
+                const zip = mod.default ?? mod;
+                read = kind === 'docx' ? await readDocx(data, file.name, maxChars, zip) : await readPptx(data, file.name, maxChars, zip);
+            }
+            if ('refusal' in read)
+                return { refusal: read.refusal };
+            out.push(new File([read.files.text.text], read.files.text.name, { type: 'text/plain' }));
+            for (const image of read.files.images)
+                out.push(new File([image.blob], image.name, { type: 'image/jpeg' }));
+        }
+        return { files: out };
+    }
     readFileAsBase64(file) {
         return new Promise((resolve, reject) => {
             const reader = new FileReader();
@@ -4242,7 +4288,7 @@ let OxAssistantChat = class OxAssistantChat extends LitElement {
       </div>
     `;
     }
-    /** The AI 작업판 card - what it holds, which version, and the way to open it. The person opens it; the model does not. */
+    /** The AI 작업 패널 card - what it holds, which version, and the way to open it. The person opens it; the model does not. */
     renderWorkspaceCard(p, key) {
         const { version, of } = versionOfCard(this.lines.map(one => one.proposals ?? []), p.workspace.name, key);
         const contents = Object.entries(p.contents ?? {})