COMSOL 6.4 - Using a Chatbot

Using a Chatbot


	The Chatbot window functionality connects to an external AI system. A subscription agreement with the AI system host is needed to obtain an API key. COMSOL is not a party in this agreement, and use of the external AI system may be subject to additional fees. It is your responsibility to ensure that your use of the AI system is in accordance with laws and policies that may apply. It is also your responsibility to review the suitability of any suggestions made by the AI system.


	The Chatbot window is only available with the Windows® version of the COMSOL Desktop. The COMSOL software must also be installed with the CHATBOT feature included in the installation and enabled using the Enable Chatbot checkbox on the Chatbot page in the Preferences window.

From the menu on the toolbar in the Windows version, choose (

) to open the window for chatting with an AI model, for example Open AI GPT-5. It can assist with generating and correcting COMSOL API for Java® code from within COMSOL Multiphysics as well as modeling questions related to the COMSOL Multiphysics software. Doing so can help out with programming using the COMSOL API for Java® as well as with general COMSOL modeling using the COMSOL Desktop. The window appears both in the Model Builder workspace and in the Application Builder workspace.


	The Chatbot window requires authentication details for your selected provider, for example, an API key or connection settings for OpenAI, Azure OpenAI, or another OpenAI API–compatible service. For more information, see www.comsol.com/learn/chatbot.

To configure a chatbot provider, click in the window. In the window, choose a provider in the list and enter a key in the field. If you are using Azure OpenAI or an OpenAI API compatible provider, see Chatbot Preference Settings for some additional settings required to set up the chatbot. Then click . You can now start a chatbot conversation.

Chatbot Conversations

From the start, the list at the top of the window only contains . In the input area at the bottom of the window (where the text appears), type a prompt (any question or statement) for the chatbot to respond to. Press Enter to submit the prompt as a message to the AI system, or click the button (

). Press Shift+Enter to add a new line to the input area for submitting prompts. The conversation is named from the first message that you submit. The reply, including your input and all further messages appear in the output area in the middle of the window. In the output area, you can right-click any message in a conversation and choose (

) (or press Ctrl+C) to copy the content of that message, or choose (

) to open the part in a window where you can see the message in a separate view. The window can be resized.

From the list, choose to start a new conversation. Then use the list to switch between all active conversations. You can also click the button (

) to add a new conversation. Click the button (

) to delete the current conversation; click the button (

) to clear all conversations.

In the input area, you can select some text and then right-click it and choose (or press Ctrl+C) to copy that text, (or press Ctrl+X) to remove it, or (or press Ctrl+V) to replace it with some previously copied text. If the cursor is outside any text, only is available, if something has previously been copied.

The messages are of the following types:

•

Messages posted by you, which have a

symbol in their upper-left corner.

•

Messages that are replies from the AI system, which have a

symbol in their upper-left corner.

•

Message that contain code, which have a

symbol in their upper-left corner.

•

Messages with an attachment, which have a

symbol.

From the list in the top-left corner of the window, you can choose from a list of subjects:

•

Choose

(the default) for any general conversations.

•

Choose

for any conversations that focus on programming using the COMSOL API for Java®.

•

Choose

for any conversations about general COMSOL modeling and the COMSOL Desktop.

Click the button (

) at the bottom of the window to attach a message with a text or image file from your file system. From the window, select a file with one of the supported formats: text, CSV, PNG, JPEG, or GIF.

Sending Attachments to the Chatbot Window

In addition to adding external files as attachments, it is possible to attach contents from a model or application to the window — for instance, Model Builder nodes, the model history, images from the Graphics window, and Application Builder methods. You can then ask the AI system for modeling advice or to assist with debugging or improving a method, for example.


	Some attachments might exceed the Message Size limit set in the Chatbot Preference Settings. If so, you will be prompted to either send the message anyway (using the number of tokens specified in the Message Size dialog) or to cancel the operation.

Model Builder Nodes

Send Model Builder nodes as attachments to the window by right-clicking one or multiple nodes in the tree and then choosing one of the following options from the submenu:

•

Choose

to send an attachment of the selected node (or nodes) to the current chatbot conversation.

•

Choose

to send an attachment of the selected node (or nodes) and all its children to the current chatbot conversation.

The attached message comprises the contents of the node as the corresponding Java code. This is an alternative to copying the node’s content using the functionality and pasting it in the input area. The difference is that when the code is sent as an attachment, only the attachment symbol (

) and the node’s name is displayed in the output area, minimizing clutter. To see the code in detail, right-click the attachment and choose .

Model History

To send the model history as an attachment to the current chatbot conversation, right-click the model’s root node and from the submenu choose (

). You will be prompted to compact the model history. In the dialog, click to attach the model history that corresponds to the current state of the model. Click to attach the entire history of the model, including settings that are no longer part of it. Right-click the attachment and choose to inspect the code in the dialog, if desired.

Graphics Window Content

Right-click in the window and choose (

) from the submenu to send a snapshot of the Graphics window as an image attachment to the current chatbot conversation. The image can be viewed by right-clicking the attachment in the window’s output area and selecting .

Application Builder Methods

When in the Application Builder, you can right-click in a method window or on a node and from the submenu choose one of the following options:

•

Choose

(for example, depending on the method name) to attach the method to a message in the window.

•

Choose

(for example, depending on the method name) to attach the method to a message in the window with a suitable prompt for the given task, which you can adjust as you see fit.

•

Choose

(for example, depending on the method name) to attach the method to a message with a suitable prompt for the given task, which you can adjust as you see fit.

These options can also be accessed by clicking in the section of the toolbar. This is an alternative to pasting the method code into the input area to ask about it if you have an Application Builder method that you have written or otherwise acquired. Right-click a method in the output area and choose (or press Ctrl+C) to copy the method content. You can also right-click and choose to remove it (or press Delete) or , to open a dialog where you can see the model content.

Connecting to the Java Shell Window

If you have code in a message in the output area (such messages have a

in the upper-left corner), then, in addition to the and options when right-clicking, you can also choose . The code is then transferred to the l window (see Interactively Using the COMSOL API for Java in the Java Shell Window), where it appears as a multiline input. Click the button (

) in the window to run the code from the window. You can also click the button (

) to send a code message to the window.

If you get an error from the suggested code when running it in the window, consider selecting the error in the window, right-click it and choose , submit it in the window, then use on the chatbot reply, run that Java code in the window, and so on until the chatbot provides correct and error-free code.

Tokens

OpenAI and Azure OpenAI uses tokens as the basis for their billing. AI language models operate on tokens that are bits of words. Both the model input and output consist of tokens and are billed for. The total tokens consumed by a conversation is defined as the sum of the input and output tokens accrued throughout a given conversation. The following two token counts are be displayed in the window’s top-right corner, next to :

•

The current conversation size, which represents the current number of tokens in the conversation.

•

Total usage throughout conversation, which represents the total number of tokens used by requests sent to the AI provider throughout the conversation.

You can see the values directly, with the current conversation size to the left and the total usage to the right, abbreviated using K or M suffixes if needed. If you hover over the numbers, a tooltip such as displays the exact token counts.


	The token counts displayed in COMSOL Multiphysics may differ from the actual usage registered by the AI provider.

Chatbot Preference Settings

The following preferences, available on the page in the window, are related to the chatbot functionality:

•

The checkbox is selected by default. Clear it to close the window and its functionality.

•

In the field, enter the threshold after which a conversation is automatically truncated, in tokens (default: 100,000 tokens).

•

In the field, enter the maximum size of an individual user message, in tokens (default: 5000 tokens). This limit on the message size only applies to user messages and not the messages related to the assistant or function calls.

•

In the field, enter the number of ongoing conversations that are stored (default: 20, upper limit: 100). When the limit is surpassed, the oldest conversations are automatically deleted. The stored conversations can be accessed from the list in the window.

•

From the list, choose the AI provider that is used for generating responses: , (the default), , or .

If you choose , also enter the following settings:

In the field, enter the endpoint URL to your Azure OpenAI service. The URL should have the format: https://comsol.openai.azure.com, where comsol is your Azure OpenAI resource name.

In the field, enter the model deployment name specified at the given Azure OpenAI endpoint.

Choose to use the window with an OpenAI API compatible AI provider. The following settings are then also required:

In the field, enter the URL to the API server. The URL has to include protocol (HTTP/HTTPS) and port number if the port is not standard.

In the field, enter the id of the language model. Reference your AI provider’s documentation for the correct values.

In the field, enter the model context size, which limits the number of input plus output tokens in any given API request (default: 128,000 tokens).

Select the checkbox to enable support for function calling. Make sure that your AI provider supports function calling, otherwise enabling this setting can cause errors.

•

The field contains the authorization key as hidden text. If authorization is not required by the provider, the field can be left blank.

•

When or is selected from the list, from the list, choose the OpenAI or Azure OpenAI model to be used by the chatbot: (the default), , , , , , or . The suffixes and indicate the reasoning effort of the GPT-5 model — is faster whereas is more thorough. If you choose , also enter the following settings for a custom model:

In the field, enter an id to determine the OpenAI or Azure OpenAI model.

In the field, enter the model context size, which limits the number of input plus output tokens in any given API request (default: 128,000 tokens).


	If you use Azure OpenAI as your provider, the model selected in the Model list must correspond to the model configuration of the Azure deployment.


	If you choose Custom from the Model list, see the OpenAI or Azure OpenAI documentation for correct values to use with the Model id and Context length (tokens) settings.