After the settings are completed, you can test the application effect in the debugging window on the right.
Standard Mode
Image QA
In Mode Setting, check the multimodal QA model, then check the image input button in Chat Experience to support image upload in the dialog box for dialogue.
Support image input in conversation for QA based on Image Content. By clicking the image icon at the bottom of the dialog box, you can upload images, call the image understanding model, and perform QA based on Image Content.
Note:
Images support jpg, png, jpeg, and bmp formats, with dimensions within 2000 pixels.
Only single image can be uploaded.
Document QA
The dialog window supports file upload for Q&A. Click the upload file icon to start the Q&A dialogue.
Note:
Supports uploading doc, docx, ppt, pptx, pdf, and txt formats.
A single file is no more than 15 MB and 1000 pages. After real-time parsing, it supports QA.
Abstract Summary Example
Support summarizing uploaded files through the model.
Document QA Example
Supporting QA based on uploaded documents.
Multi-Agent Mode
Image Upload
Note:
Images support .jpg, .jpeg, .png, .bmp, and webp formats, with dimensions within 2000 pixels.
Supports uploading up to 20 images.
Allows uploading does not mean the Agent can process image. The Agent can correctly process image content only when dependent on corresponding configuration plug-in.
Allows uploading does not mean the Agent can process the file. The Agent can correctly handle file content only when dependent on corresponding configuration plug-in.
General Settings
Debugging Parameters
Click the debugging parameters icon to set custom API parameter values during debugging, making it easy to perform feature debugging.
Preset the VisitorBizId parameter, which can be filled in for unique identification to tag the current session user integration.
You can also add custom API parameters in app settings >variables and memory. For detailed parameter description, please see application-level variables.
Comparing Debugging
In standard mode and Multi-Agent Mode (single Agent), it supports compare debugging for multiple models/prompt content.
Support simultaneous creation of up to 3 comparison groups for comparing response performance of different models or prompt content under identical questions. Each comparison group supports:
Select model and adjust related parameters
Edit prompt content, use Version Used/Template/AI one-click optimization feature to optimize prompt content
Select the better-performing comparison group based on comparison results, click "Select this" to select. When exiting the comparison:
If you select "save and exit", the currently selected model and prompt content will apply to projects.
If you select "Discard and exit", any changes will not be saved.
Reference Source
After enabling the Show Reference Source toggle in document settings, answer reference video clips will be shown at the bottom of the reply.
Simultaneously support tracing the source to the original document, support viewing the source file in the dialogue test webpage, and navigate to the page where the corresponding fragment resides in a new tag page (support pdf\\doc\\ppt\\pptx); csv\\xlsx\\xls support redirection to specific sheet, and images support turning on the original image.
When the input question used a search engine to retrieve an answer online, the response reference external link will be displayed and shown as a superscript in the text.
Search engine configuration method:
1. Click Settings for online search. The international site currently supports interfacing with Tavily to perform online search. You need to get a Tavily API Key.
2. Once configured, toggle on the online search switch to proceed with online QA.
3. Dialogue example show
Debug Information
The dialogue debug entry is located at the bottom of the dialogue message. After clicking, it shows debug details.
Debug details include the following three aspects:
2. Dialogue call tree: It shows the complete response process of this dialogue request in a tree structure, making it easy to view the sequence of steps and branch logic during operation, and determine whether it complies with expectations (for example: whether tools are mistakenly called or knowledge library search is missed).
3. Running details of steps in the dialogue call tree:
Take the "call workflow" step as an example: click a specific node to view the request status, overall duration, token consumption, and input/output details of this step.
Click Workflow > Running details in the call workflow to view the complete workflow canvas info. Click the node detail in the workflow canvas, and a popup displays the input and output variables of the node.
Submit Feedback
On the application configuration page, you can view the call information and detect errors in the model reply. Click Submit Feedback in the bottom-right corner of the dialog box.
Backend operations staff will troubleshoot based on the information you provide. You can check the feedback results in User Profile Photo > My Feedback.
Regenerate
Click Regenerate, the model will resend this round's question and generate the answer again.
Revise
Click Reviseto support entering the current Q&A into the Q&A database. By manually correcting the answer, you can fix the current issue reply. Next time a similar type of question occurs, the system will preferentially match the corrected answer from the Q&A database.
