1 files changed, 22 insertions, 2 deletions

diff --git a/src/fields/Doc.ts b/src/fields/Doc.ts
index e62ca4bb8..950d9047c 100644
--- a/src/fields/Doc.ts
+++ b/src/fields/Doc.ts
@@ -7,7 +7,7 @@ import { CollectionViewType, DocumentType } from '../client/documents/DocumentTy
 import { scriptingGlobal, ScriptingGlobals } from '../client/util/ScriptingGlobals';
 import { afterDocDeserialize, autoObject, Deserializable, SerializationHelper } from '../client/util/SerializationHelper';
 import { undoable, UndoManager } from '../client/util/UndoManager';
-import { ClientUtils, incrementTitleCopy } from '../ClientUtils';
+import { ClientUtils, imageUrlToBase64, incrementTitleCopy } from '../ClientUtils';
 import {
     AclAdmin, AclAugment, AclEdit, AclPrivate, AclReadonly, Animation, AudioPlay, Brushed, CachedUpdates, DirectLinks,
     DocAcl, DocCss, DocData, DocLayout, DocViews, FieldKeys, FieldTuples, ForceServerWrite, Height, Highlight,
@@ -22,8 +22,9 @@ import { FieldId, RefField } from './RefField';
 import { RichTextField } from './RichTextField';
 import { listSpec } from './Schema';
 import { ComputedField, ScriptField } from './ScriptField';
-import { BoolCast, Cast, DocCast, FieldValue, NumCast, StrCast, ToConstructor, toList } from './Types';
+import { BoolCast, Cast, DocCast, FieldValue, ImageCastWithSuffix, NumCast, RTFCast, StrCast, ToConstructor, toList } from './Types';
 import { containedFieldChangedHandler, deleteProperty, GetEffectiveAcl, getField, getter, makeEditable, makeReadOnly, setter, SharingPermissions } from './util';
+import { gptImageLabel } from '../client/apis/gpt/GPT';
 
 export let ObjGetRefField: (id: string, force?: boolean) => Promise<Doc | undefined>;
 export let ObjGetRefFields: (ids: string[]) => Promise<Map<string, Doc | undefined>>;
@@ -1467,6 +1468,25 @@ export namespace Doc {
         });
     }
 
+    /**
+     * text description of a Doc. RTF documents will have just their text and pdf documents will have the first 50 words.
+     * Image documents are converted to bse64 and gpt generates a description for them. all other documents use their title.
+     * @param doc
+     * @returns
+     */
+    export function getDescription(doc: Doc) {
+        const curDescription = StrCast(doc[DocData][Doc.LayoutFieldKey(doc) + '_description']);
+        const docText = (async (tdoc:Doc) => {
+            switch (tdoc.type) {
+                case DocumentType.PDF: return curDescription || StrCast(tdoc.text).split(/\s+/).slice(0, 50).join(' '); // first 50 words of pdf text
+                case DocumentType.IMG: return curDescription || imageUrlToBase64(ImageCastWithSuffix(Doc.LayoutField(tdoc), '_o') ?? '')
+                                                .then(hrefBase64 => gptImageLabel(hrefBase64, 'Give three to five labels to describe this image.'));
+                case DocumentType.RTF: return RTFCast(tdoc[Doc.LayoutFieldKey(tdoc)]).Text;
+                default:               return StrCast(tdoc.title);
+        }}); // prettier-ignore
+        return docText(doc).then(text => (doc[DocData][Doc.LayoutFieldKey(doc) + '_description'] = text));
+    }
+
     // prettier-ignore
     export function toIcon(doc?: Doc, isOpen?: Opt<boolean>) {
         if (isOpen) return doc?.isFolder ? 'chevron-down' : 'folder-open';