Link draw.io and ChatGPT

[alderg]

You can now enable a generic ChatGPT window which allows to create new diagrams, ask questions about the current diagram, selection or the app (help). The feature is experimental and requires a ChatGPT API key. To enable the feature, use gpt-api-key=[your-api-key] as an URL parameter or specify gptApiKey in the configuration (see below). An optional gpt-model=[model-name] can also be specified. Possible values are gpt-4 or gpt-3.5-turbo (default), depending on the API key.

If an ChatGPT API key is present, the Chat menu will be shown under the Tags menu:

Examples

Creating diagrams and summaries:

Asking questions about cross-functional flowchart roles:

Changing styles of vertices and edges:

Describing and interpreting org charts:

Asking questions about class diagrams and multiplicities:

Describing sequence diagrams:

Limitations for number of tokens on large diagrams:

Asking questions about application functionality (help):

[JustSaX]

Is a specific version necessary? I'm using v23.0.2 on mac. An entered the following in configuration:

{ "gptApiKey": "xxxx" }

I don't have the option for chat in the view menu.

[davidjgraph]

the offline desktop app?

[JustSaX]

yes

[davidjgraph]

it's offline, meaning it doesn't connect to any online servers, as would be required to call chatgpt.

[thecopeeninja]

Can we use self hosted LLM models in configuration file?

{
  "enableChatGpt": true,
  "gptApiKey": "dummy-key",
  "gptModel": "TechxGenus/DeepSeek-Coder-V2-Lite-Instruct-AWQ",
  "gptUrl": "http://YOUR_TAILSCALE_GPU_IP:8000/v1/chat/completions"
}

[alderg]

It should work.

[thecopeeninja]

I have been trying for the past two weeks but I keep getting error in chat interface when using above custom configuration.
Are you able to replicate it on your end?

Also, the windows installed app of drawio editor doesn't support chat interface. Only web browser version supports it for now.

[davidjgraph]

1. What is returned in the network tab of the browser debugger for that call?
2. That statement is correct.

[hushoujier]

May I ask if it is possible to add a Content-Security-Policy for https://api.siliconflow.cn in content-src。

Thank you.

[davidjgraph]

Absoutely not.

[futugyou]

May I ask if it is possible to add a Content-Security-Policy for https://models.github.ai in content-src。
Same reason as the previous reply.

[davidjgraph]

Absolutely not. If you want to allow XHR to some random site, host the project yourself and allow the CSP.

[futugyou]

Can't you see that this is provided by github, and definitely not the so-called 'some random site' you mentioned.
Your CSP list has https://api.github.com, which shows that you are not excluding github, and https://models.github.ai is provided by github.

[davidjgraph]

I don't care. We don't use the domain, we're not adding it.

[tfbroner]

Can't you see that this is provided by github, and definitely not the so-called 'some random site' you mentioned. Your CSP list has https://api.github.com, which shows that you are not excluding github, and https://models.github.ai is provided by github.

You can't be serious. How can you ask JGraph to vouch for the content served on a remote server which is NOT under JGraph's control? It's absurd.

[mesbahworld]

Because Open AI are not free and I want to add some opensource or free option but unable to get succeed.

For Gemini they don't use bearer in the header and it seems drawio by default using bearer and that's why i am getting error.
If i use Ollama and for test I did not use any authentication. so if i dont use gptApiKey the chat option not appear and if i use dummy key then the request failed.

Gemini:

curl --location 'https://generativelanguage.googleapis.com/v1beta/models/gemini-2.0-flash:generateContent' \
--header 'Content-Type: application/json' \
--header 'X-goog-api-key: MyG00g1e_T0KeN' \
--data '{
    "contents": [
      {
        "parts": [
          {
            "text": "Explain how AI works in a few words"
          }
        ]
      }
    ]
  }'

Ollama:

curl --location 'http://MY_IP:11434/api/chat' \
--header 'Content-Type: application/json' \
--data '{
    "model": "codellama:13b",
    "messages": [
      {
        "role": "user",
        "content": "Why is the sky blue?"
      }
    ]
  }'
  

[mesbahworld]

Adding Custom header option would be nice to have other vendor
or,
Let user choose the full 'key : value' as user input instead of default 'bearer' : key_user_input