ChatDOC Studio Quick Start Guide

ChatDOC Studio allows you to create and launch AI applications in four simple steps: Sources, Configure, Test, and Launch.

1. Sources: Import data from any source—PDFs, Word documents, websites, and more.
2. Configure: Choose your AI model and customize its personality.
3. Test: Ask questions at the interactive playground to test responses.
4. Launch: Integrate the app on your website.

I. Creating and Launching a Chat App

The Chat App acts as your rigorous AI help assistant, grounded entirely in your files, and is built for 24/7 external customer service or internal assistance.

Step 1: Create the App and Select Style

1.Log in to ChatDoc Studio.

2.Locate the Chat App and click the “Create” button.

3.Choose the Chat App style you wish to create:

• Floating Chat Component: An easy option for quick checks, allowing users to ask questions in a window floating above the document without distraction.
• Panoramic Embedding: An integrated view suitable for scenarios like research and verification, allowing direct comparison between the AI's answer and the source text.

4.After creation, you will enter the Playground interface, where testing and parameter settings can be managed.

tep 2: Import Sources and Train the AI

1.Navigate to the Source interface. The Source page manages all content sources supporting your Chat AI App, ensuring accurate, relevant, and current information.

2.Select your data sources for training. Current supported methods are:

• File: Upload and manage various document types, including PDF, DOC, DOCX, MD, and TXT. Notably, only PDF, DOC, and DOCX formats support answer source tracing.
• Website: Crawl content from a URL you input.
• Knowledge: Select files already added to your team's Knowledge Hub. Note: Any file uploaded by any user within the team for any App is automatically added to the Knowledge Hub.

3.If no source is selected, the chat will interact directly with the Large Language Model (LLM).

4.Files enter a processing queue after upload and are displayed in the "Linked" section, allowing you to view the upload progress.

5.In the Sources module, train the App. After changing the data source scope, re-training is necessary for the changes to take effect.

Step 3: Configure and Test the App (Playground)

Use the Playground page to interact with and debug your Chat App to ensure it meets your requirements before deployment.

1.Configure General Settings:

• Name: Customize the Chat App's name.
• Logo: Customize the avatar for the Q&A responses.
• Temperature: Control the randomness of the AI response (lower values result in more focused answers).
• Instruction: Send a custom prompt to the AI model to guide the response style and content.

2.Configure Interface Settings:

• Welcome message: Customize the automatic message sent to users when they start a conversation.
• Input placeholder: Customize the prompt text displayed inside the input box.
• Answer Source Tracing: Toggle to enable/disable fine-grained source tracing, which supports document preview and allows users to trace any selected answer segment back to the relevant source text snippet (supported for PDF, DOC, and DOCX files).
• Support new chat/chat history: Control whether to allow users to start a new dialogue or view historical conversations.
• Predominant color of the interface: Customize the theme color of the chat application interface to match your brand.
• Floating icon/Position: Customize the color and set the alignment (left or right) of the floating chat icon.

Step 4: Launch and Integrate the App

Once the Chat App is debugged, you can choose how to integrate it so users can interact with it.
You have four integration options:

1.Embed the floating button/Embed: Add a floating button to a specific website page that opens a floating Q&A dialogue window upon clicking. This ensures the Chat App does not interfere with the user's browsing. Example: Paste the provided code snippet into the website's section or elsewhere on the page.

2.Embed the iFrame directly/IFrame: Seamlessly embed the Q&A dialogue window onto a full-screen page, where it remains open and accessible.

3.Publicly accessible URL link/URL: Provide a publicly accessible link that takes users directly to the full-screen Q&A dialogue window. This link can be embedded in emails or social media.

4.API interface/API: Integrate the Chat App with any business system using API calls, allowing you to design a fully customized Q&A experience that aligns with your brand.
After integration, you will have a running Chat AI App that is live and available to assist users.

II. Guide for Context Retrieval App

The Context Retrieval App uses a system based on context and document outline awareness to find the most relevant information within your data sources.

1.Create: Log in, find the Context Retrieval App, and click "Create".

2.Source: Select data sources (File or Knowledge) for training. Supported file formats include PDF, DOC, DOCX, MD, and TXT (PDF/DOC/DOCX support tracing).

3.Train: Re-training is required after changes to the data source scope.

4.Configure/Test (Playground): Use the Playground to interact and debug the Content Retrieval results. Parameters available include:

• Retrieval Result Token Length: Controls the length of the returned text.
• Basic (Retrieval → Non-contextual Rerank):
  Fast and efficient. Combines Embedding and BM25 hybrid retrieval, followed by a non-contextual reranker to reorder the results.
• Contextual (Retrieval → Contextual Rerank):
  More precise. Combines Embedding and BM25 hybrid retrieval, followed by a contextual reranker to reorder the results for better accuracy.
• Expanded (Retrieval → Contextual Rerank → Context Expansion → Contextual Rerank):
  More comprehensive and highly accurate, with increased latency. Following the initial contextual reranking, additional context from the most relevant segments is incorporated into the candidate set for a subsequent reranking phase.
• Doc Outline-aware Retrieval: Toggles whether to integrate the document’s directory structure into the retrieval process, suitable for documents with complex hierarchy.

5.Launch: Integration is achieved exclusively through API calls.

III. Guide for PDF Parser App

The PDF Parser App converts uploaded PDFs into structured, LLM-ready data while preserving the original formatting and layout.

1.Access: Log in, find the PDF Parser App, and click the “Playground” button.

2.Source: In the Playground, select the PDF document you want to parse (File or Knowledge).

3.Process: When you upload a document, it’s parsed automatically and a structured result is generated. You can also view the history of parsed documents.

4.Launch: Integration is achieved through API.