@makemind/sdk 0.2.0 → 0.2.1

package/CHANGELOG.md

@@ -5,6 +5,21 @@ All notable changes to `@makemind/sdk` will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## 0.2.1 — 2026-10-12
+
+### Fixed
+
+- `ai.analyzeDocument()` sends the document. It used to send the words
+  "Document URL: …" to a chat model, which never saw the document and answered
+  anyway. The document goes as a `document` part (an https address or a data
+  URI), and only a model that reads PDFs is chosen for it.
+
+### Added
+
+- `ai.ocr(document, { pages?, mediaType?, model? })` — the document's text as
+  written, page by page and line by line with confidence and position; at most
+  30 pages per call, `more` while a next window may hold pages.
+
 ## 0.2.0 — 2026-10-09
 
 ### Changed — breaking

package/dist/index.d.mts

@@ -1907,6 +1907,30 @@ interface ToolDefinition {
     description: string;
     parameters: Record<string, unknown>;
 }
+/** One line an OCR engine read: its text, confidence and box (0–1: left, top, right, bottom). */
+interface OcrLine {
+    text: string;
+    confidence: number;
+    box: [number, number, number, number];
+}
+/** One page an OCR engine read. */
+interface OcrPage {
+    number: number;
+    width: number;
+    height: number;
+    text: string;
+    confidence: number;
+    lines: OcrLine[];
+}
+/** What `ocr()` answers: the pages read, and whether a next window may hold more. */
+interface OcrResult {
+    requestId: string;
+    model: string;
+    pages: OcrPage[];
+    more: boolean;
+    cost: number;
+    currency?: string;
+}
 declare class MAKEMINDAI {
     private apiClient;
     private getTenantId;
@@ -1991,20 +2015,35 @@ declare class MAKEMINDAI {
     }>;
     /**
      * Document analysis (summary, field extraction, or QA).
-     * No dedicated onCall exists in the ai domain; implemented via ai-aiChat
-     * with a structured system prompt that targets the requested analysis type.
-     * The document content must be fetched by the caller and passed as text —
-     * the backend has no URL-fetch / OCR pipeline for onCall yet.
+     * The document itself is sent — a PDF as a `document` part, which the
+     * platform hands only to a model that reads PDFs (core-ai 18 §5.2). `document`
+     * is an https address the platform reads, or a `data:application/pdf;base64,…`
+     * URI.
      */
-    analyzeDocument(documentUrl: string, analysisType: 'summary' | 'extract' | 'qa', options?: {
+    analyzeDocument(document: string, analysisType: 'summary' | 'extract' | 'qa', options?: {
         question?: string;
         extractFields?: string[];
+        model?: string;
     }): Promise<{
         result: string | Record<string, unknown>;
         usage: {
             totalTokens: number;
         };
     }>;
+    /**
+     * A document's text as written, page by page and line by line, from an OCR
+     * engine — each line with its confidence and its box on the page (core-ai 18
+     * §5.3). One call reads at most 30 pages; read the next window while `more`.
+     * `document` is an https address or a data URI (PDF, PNG, JPEG, TIFF).
+     */
+    ocr(document: string, options?: {
+        pages?: {
+            from?: number;
+            to?: number;
+        };
+        mediaType?: string;
+        model?: string;
+    }): Promise<OcrResult>;
     generate(prompt: string, options?: ChatOptions): Promise<{
         text: string;
         usage: {
@@ -3083,4 +3122,4 @@ declare const ErrorCodes: {
     readonly UNKNOWN_ERROR: "UNKNOWN_ERROR";
 };
 
-export { ApiClient, type ApiClientConfig, type ApiRequestOptions, type AuthStateCallback, type AuthUser, type BillingSummary, type CollectionSnapshotCallback, type ConsultResult, type ConsultationConclusion, type CreateOptions, type DeleteOptions, type DocumentResult, ErrorCodes, type ExpertAccess, ExpertChat, type ExpertChatWidget, type ExpertChatWidgetOptions, type ExpertConsultation, ExpertDoorError, type ExpertSession, type ExpertTurn, type FileMetadata, type FlowListItem, type GetOptions, type Invoice, type ListFilesOptions, type ListFilesResult, type ListFlowsParams, type ListFlowsResult, type LogicCallResult, MAKEMINDAI, type MAKEMINDAIConfig, MAKEMINDApps, type MAKEMINDAppsConfig, MAKEMINDAuth, type MAKEMINDAuthConfig, MAKEMINDClient, type MAKEMINDClientConfig, type MAKEMINDConfig, MAKEMINDData, type MAKEMINDDataConfig, MAKEMINDError, MAKEMINDExperts, type MAKEMINDExpertsConfig, MAKEMINDLogic, type MAKEMINDLogicConfig, MAKEMINDMarketplace, type MAKEMINDMarketplaceConfig, MAKEMINDMessaging, type MAKEMINDMessagingConfig, MAKEMINDSearch, type MAKEMINDSearchConfig, MAKEMINDSettlement, type MAKEMINDSettlementConfig, MAKEMINDStorage, type MAKEMINDStorageConfig, MAKEMINDTenant, type MAKEMINDTenantConfig, type Notification, type OrderBy, type PlatformErrorDetails, type QueryCondition, type QueryOptions, type QueryResult, type SendMessageOptions, type SnapshotCallback, type Tenant, type TenantClaims, type TenantMember, type Unsubscribe, type UpdateOptions, type UploadOptions, type UploadResult, type UsageEventOptions, type UsageStats, type WorkflowTransitionOptions, type Workspace, getMAKEMIND, initializeMAKEMIND, mountExpertChat };
+export { ApiClient, type ApiClientConfig, type ApiRequestOptions, type AuthStateCallback, type AuthUser, type BillingSummary, type CollectionSnapshotCallback, type ConsultResult, type ConsultationConclusion, type CreateOptions, type DeleteOptions, type DocumentResult, ErrorCodes, type ExpertAccess, ExpertChat, type ExpertChatWidget, type ExpertChatWidgetOptions, type ExpertConsultation, ExpertDoorError, type ExpertSession, type ExpertTurn, type FileMetadata, type FlowListItem, type GetOptions, type Invoice, type ListFilesOptions, type ListFilesResult, type ListFlowsParams, type ListFlowsResult, type LogicCallResult, MAKEMINDAI, type MAKEMINDAIConfig, MAKEMINDApps, type MAKEMINDAppsConfig, MAKEMINDAuth, type MAKEMINDAuthConfig, MAKEMINDClient, type MAKEMINDClientConfig, type MAKEMINDConfig, MAKEMINDData, type MAKEMINDDataConfig, MAKEMINDError, MAKEMINDExperts, type MAKEMINDExpertsConfig, MAKEMINDLogic, type MAKEMINDLogicConfig, MAKEMINDMarketplace, type MAKEMINDMarketplaceConfig, MAKEMINDMessaging, type MAKEMINDMessagingConfig, MAKEMINDSearch, type MAKEMINDSearchConfig, MAKEMINDSettlement, type MAKEMINDSettlementConfig, MAKEMINDStorage, type MAKEMINDStorageConfig, MAKEMINDTenant, type MAKEMINDTenantConfig, type Notification, type OcrLine, type OcrPage, type OcrResult, type OrderBy, type PlatformErrorDetails, type QueryCondition, type QueryOptions, type QueryResult, type SendMessageOptions, type SnapshotCallback, type Tenant, type TenantClaims, type TenantMember, type Unsubscribe, type UpdateOptions, type UploadOptions, type UploadResult, type UsageEventOptions, type UsageStats, type WorkflowTransitionOptions, type Workspace, getMAKEMIND, initializeMAKEMIND, mountExpertChat };

package/dist/index.d.ts

@@ -1907,6 +1907,30 @@ interface ToolDefinition {
     description: string;
     parameters: Record<string, unknown>;
 }
+/** One line an OCR engine read: its text, confidence and box (0–1: left, top, right, bottom). */
+interface OcrLine {
+    text: string;
+    confidence: number;
+    box: [number, number, number, number];
+}
+/** One page an OCR engine read. */
+interface OcrPage {
+    number: number;
+    width: number;
+    height: number;
+    text: string;
+    confidence: number;
+    lines: OcrLine[];
+}
+/** What `ocr()` answers: the pages read, and whether a next window may hold more. */
+interface OcrResult {
+    requestId: string;
+    model: string;
+    pages: OcrPage[];
+    more: boolean;
+    cost: number;
+    currency?: string;
+}
 declare class MAKEMINDAI {
     private apiClient;
     private getTenantId;
@@ -1991,20 +2015,35 @@ declare class MAKEMINDAI {
     }>;
     /**
      * Document analysis (summary, field extraction, or QA).
-     * No dedicated onCall exists in the ai domain; implemented via ai-aiChat
-     * with a structured system prompt that targets the requested analysis type.
-     * The document content must be fetched by the caller and passed as text —
-     * the backend has no URL-fetch / OCR pipeline for onCall yet.
+     * The document itself is sent — a PDF as a `document` part, which the
+     * platform hands only to a model that reads PDFs (core-ai 18 §5.2). `document`
+     * is an https address the platform reads, or a `data:application/pdf;base64,…`
+     * URI.
      */
-    analyzeDocument(documentUrl: string, analysisType: 'summary' | 'extract' | 'qa', options?: {
+    analyzeDocument(document: string, analysisType: 'summary' | 'extract' | 'qa', options?: {
         question?: string;
         extractFields?: string[];
+        model?: string;
     }): Promise<{
         result: string | Record<string, unknown>;
         usage: {
             totalTokens: number;
         };
     }>;
+    /**
+     * A document's text as written, page by page and line by line, from an OCR
+     * engine — each line with its confidence and its box on the page (core-ai 18
+     * §5.3). One call reads at most 30 pages; read the next window while `more`.
+     * `document` is an https address or a data URI (PDF, PNG, JPEG, TIFF).
+     */
+    ocr(document: string, options?: {
+        pages?: {
+            from?: number;
+            to?: number;
+        };
+        mediaType?: string;
+        model?: string;
+    }): Promise<OcrResult>;
     generate(prompt: string, options?: ChatOptions): Promise<{
         text: string;
         usage: {
@@ -3083,4 +3122,4 @@ declare const ErrorCodes: {
     readonly UNKNOWN_ERROR: "UNKNOWN_ERROR";
 };
 
-export { ApiClient, type ApiClientConfig, type ApiRequestOptions, type AuthStateCallback, type AuthUser, type BillingSummary, type CollectionSnapshotCallback, type ConsultResult, type ConsultationConclusion, type CreateOptions, type DeleteOptions, type DocumentResult, ErrorCodes, type ExpertAccess, ExpertChat, type ExpertChatWidget, type ExpertChatWidgetOptions, type ExpertConsultation, ExpertDoorError, type ExpertSession, type ExpertTurn, type FileMetadata, type FlowListItem, type GetOptions, type Invoice, type ListFilesOptions, type ListFilesResult, type ListFlowsParams, type ListFlowsResult, type LogicCallResult, MAKEMINDAI, type MAKEMINDAIConfig, MAKEMINDApps, type MAKEMINDAppsConfig, MAKEMINDAuth, type MAKEMINDAuthConfig, MAKEMINDClient, type MAKEMINDClientConfig, type MAKEMINDConfig, MAKEMINDData, type MAKEMINDDataConfig, MAKEMINDError, MAKEMINDExperts, type MAKEMINDExpertsConfig, MAKEMINDLogic, type MAKEMINDLogicConfig, MAKEMINDMarketplace, type MAKEMINDMarketplaceConfig, MAKEMINDMessaging, type MAKEMINDMessagingConfig, MAKEMINDSearch, type MAKEMINDSearchConfig, MAKEMINDSettlement, type MAKEMINDSettlementConfig, MAKEMINDStorage, type MAKEMINDStorageConfig, MAKEMINDTenant, type MAKEMINDTenantConfig, type Notification, type OrderBy, type PlatformErrorDetails, type QueryCondition, type QueryOptions, type QueryResult, type SendMessageOptions, type SnapshotCallback, type Tenant, type TenantClaims, type TenantMember, type Unsubscribe, type UpdateOptions, type UploadOptions, type UploadResult, type UsageEventOptions, type UsageStats, type WorkflowTransitionOptions, type Workspace, getMAKEMIND, initializeMAKEMIND, mountExpertChat };
+export { ApiClient, type ApiClientConfig, type ApiRequestOptions, type AuthStateCallback, type AuthUser, type BillingSummary, type CollectionSnapshotCallback, type ConsultResult, type ConsultationConclusion, type CreateOptions, type DeleteOptions, type DocumentResult, ErrorCodes, type ExpertAccess, ExpertChat, type ExpertChatWidget, type ExpertChatWidgetOptions, type ExpertConsultation, ExpertDoorError, type ExpertSession, type ExpertTurn, type FileMetadata, type FlowListItem, type GetOptions, type Invoice, type ListFilesOptions, type ListFilesResult, type ListFlowsParams, type ListFlowsResult, type LogicCallResult, MAKEMINDAI, type MAKEMINDAIConfig, MAKEMINDApps, type MAKEMINDAppsConfig, MAKEMINDAuth, type MAKEMINDAuthConfig, MAKEMINDClient, type MAKEMINDClientConfig, type MAKEMINDConfig, MAKEMINDData, type MAKEMINDDataConfig, MAKEMINDError, MAKEMINDExperts, type MAKEMINDExpertsConfig, MAKEMINDLogic, type MAKEMINDLogicConfig, MAKEMINDMarketplace, type MAKEMINDMarketplaceConfig, MAKEMINDMessaging, type MAKEMINDMessagingConfig, MAKEMINDSearch, type MAKEMINDSearchConfig, MAKEMINDSettlement, type MAKEMINDSettlementConfig, MAKEMINDStorage, type MAKEMINDStorageConfig, MAKEMINDTenant, type MAKEMINDTenantConfig, type Notification, type OcrLine, type OcrPage, type OcrResult, type OrderBy, type PlatformErrorDetails, type QueryCondition, type QueryOptions, type QueryResult, type SendMessageOptions, type SnapshotCallback, type Tenant, type TenantClaims, type TenantMember, type Unsubscribe, type UpdateOptions, type UploadOptions, type UploadResult, type UsageEventOptions, type UsageStats, type WorkflowTransitionOptions, type Workspace, getMAKEMIND, initializeMAKEMIND, mountExpertChat };

package/dist/index.js

@@ -2851,12 +2851,12 @@ var MAKEMINDAI = class {
   }
   /**
    * Document analysis (summary, field extraction, or QA).
-   * No dedicated onCall exists in the ai domain; implemented via ai-aiChat
-   * with a structured system prompt that targets the requested analysis type.
-   * The document content must be fetched by the caller and passed as text —
-   * the backend has no URL-fetch / OCR pipeline for onCall yet.
+   * The document itself is sent — a PDF as a `document` part, which the
+   * platform hands only to a model that reads PDFs (core-ai 18 §5.2). `document`
+   * is an https address the platform reads, or a `data:application/pdf;base64,…`
+   * URI.
    */
-  async analyzeDocument(documentUrl, analysisType, options) {
+  async analyzeDocument(document2, analysisType, options) {
     const systemPromptMap = {
       summary: "You are a document summarization assistant. Summarize the provided document concisely.",
       extract: `You are a document extraction assistant. Extract the following fields from the document as JSON: ${(options?.extractFields ?? []).join(", ")}.`,
@@ -2864,9 +2864,10 @@ var MAKEMINDAI = class {
     };
     const response = await this.apiClient.dispatch("ai-aiChat", {
       tenantId: this.getTenantId(),
+      ...options?.model ? { model: options.model } : {},
       messages: [
         { role: "system", content: systemPromptMap[analysisType] },
-        { role: "user", content: `Document URL: ${documentUrl}` }
+        { role: "user", content: [{ type: "document", document: { url: document2 } }] }
       ]
     });
     let result = response.content;
@@ -2882,6 +2883,20 @@ var MAKEMINDAI = class {
       usage: { totalTokens: response.usage.totalTokens }
     };
   }
+  /**
+   * A document's text as written, page by page and line by line, from an OCR
+   * engine — each line with its confidence and its box on the page (core-ai 18
+   * §5.3). One call reads at most 30 pages; read the next window while `more`.
+   * `document` is an https address or a data URI (PDF, PNG, JPEG, TIFF).
+   */
+  async ocr(document2, options) {
+    return this.apiClient.dispatch("ai-aiOcr", {
+      tenantId: this.getTenantId(),
+      document: { url: document2, ...options?.mediaType ? { mediaType: options.mediaType } : {} },
+      ...options?.pages ? { pages: options.pages } : {},
+      ...options?.model ? { model: options.model } : {}
+    });
+  }
   // Text generation
   async generate(prompt, options) {
     const response = await this.chat(

package/dist/index.mjs

@@ -2795,12 +2795,12 @@ var MAKEMINDAI = class {
   }
   /**
    * Document analysis (summary, field extraction, or QA).
-   * No dedicated onCall exists in the ai domain; implemented via ai-aiChat
-   * with a structured system prompt that targets the requested analysis type.
-   * The document content must be fetched by the caller and passed as text —
-   * the backend has no URL-fetch / OCR pipeline for onCall yet.
+   * The document itself is sent — a PDF as a `document` part, which the
+   * platform hands only to a model that reads PDFs (core-ai 18 §5.2). `document`
+   * is an https address the platform reads, or a `data:application/pdf;base64,…`
+   * URI.
    */
-  async analyzeDocument(documentUrl, analysisType, options) {
+  async analyzeDocument(document2, analysisType, options) {
     const systemPromptMap = {
       summary: "You are a document summarization assistant. Summarize the provided document concisely.",
       extract: `You are a document extraction assistant. Extract the following fields from the document as JSON: ${(options?.extractFields ?? []).join(", ")}.`,
@@ -2808,9 +2808,10 @@ var MAKEMINDAI = class {
     };
     const response = await this.apiClient.dispatch("ai-aiChat", {
       tenantId: this.getTenantId(),
+      ...options?.model ? { model: options.model } : {},
       messages: [
         { role: "system", content: systemPromptMap[analysisType] },
-        { role: "user", content: `Document URL: ${documentUrl}` }
+        { role: "user", content: [{ type: "document", document: { url: document2 } }] }
       ]
     });
     let result = response.content;
@@ -2826,6 +2827,20 @@ var MAKEMINDAI = class {
       usage: { totalTokens: response.usage.totalTokens }
     };
   }
+  /**
+   * A document's text as written, page by page and line by line, from an OCR
+   * engine — each line with its confidence and its box on the page (core-ai 18
+   * §5.3). One call reads at most 30 pages; read the next window while `more`.
+   * `document` is an https address or a data URI (PDF, PNG, JPEG, TIFF).
+   */
+  async ocr(document2, options) {
+    return this.apiClient.dispatch("ai-aiOcr", {
+      tenantId: this.getTenantId(),
+      document: { url: document2, ...options?.mediaType ? { mediaType: options.mediaType } : {} },
+      ...options?.pages ? { pages: options.pages } : {},
+      ...options?.model ? { model: options.model } : {}
+    });
+  }
   // Text generation
   async generate(prompt, options) {
     const response = await this.chat(

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@makemind/sdk",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "description": "MAKEMIND Platform SDK for Web Applications",
   "main": "dist/index.js",
   "module": "dist/index.mjs",