Using Custom LLM Backends in draw.io

[alderg]

This topic has moved to https://www.drawio.com/doc/faq/configure-ai-options

[thecopeeninja]

Thank you for making it standardized across GPTs like Claude, OpenAI, Gemini, self hosted etc.

By when can we expect the release in windows drawio editor?

PS: I tried adding the config but the generate button doesn't show up in editor.

[davidjgraph]

This is live in the 29.2.9 release.

[mradulgautam01]

Is this version available to run on desktop? @davidjgraph can you please attach the binaries
If yes, please share the link as the latest version I can find is 29.0.3 for desktop

[davidjgraph]

No, the desktop app is isolated by design.

[luxinming]

I tried adding the config but the generate button doesn't show up in editor.
29.3.0 for web

[alderg]

What have you tried?

[kdwhan]

Hi, i'm using the drawio plugin 14.2.7 in Confluence (data center).

1. Is there a plugin version that supports this feature? If not, when will it become available?
2. Is it possible to configure custom LLM settings in the Confluence plugin?

[sahil-datascience]

I do not see any options to set Model configuration, is this feature disabled in the current version? Also, are there any chances of Generate option coming to desktop app?

[davidjgraph]

No, the desktop app is isolated by design.

[ralphilius]

I don't see dropdowns to select actions and models.. Do I need to add them in configuration?

[alderg]

tl;dr try this via Extras, Configuration:

{
  "gptApiKey": "sk-proj-XXXX",
  "geminiApiKey": "XXXX",
  "claudeApiKey": "sk-ant-api03-XXXX"
}

[pjae26]

sorry can anybody please explain how this options are to use within the online version if the desktop version (the one i can fork from github source code) has no integrations?
how to roll this out in an enterprise environment?
thanks

[alderg]

Generate a key and put that in the above configuration via Extras, Configuration.

[bulme-sm]

I am trying to get the AI feature working with V29.6.6 downloaded as standalone version (free) on macOS with OpenRouter.ai
OpenRouter.ai is an OpenAI API compatible Proxy for different LLMs.

My entry in Extras, Configuration:

{
"enableAi": true,
"gptUrl": "https://openrouter.ai/api/v1/chat/completions",
"gptApiKey": "sk-or-v1-xxxx-my-key-xxxxxxxxxxxx"
}

I even enabled logging via
/Applications/draw.io.app/Contents/MacOS/draw.io --enable-logging

But the sparkle button doesn't show up and I don't see any error or warning in logging regarding this.

I also saw the documentation here: https://www.drawio.com/doc/faq/configure-ai-options
and it looks that there are several more config options, but I didn't find a minimal working example.
Is it even possible in my version? What I am missing? Could anyone give us a minimal working config and possible an enhanced one?

[alderg]

#5387 (reply in thread)

[bulme-sm]

Thanks for your reply. So not possible in the Desktop version...
Is there a way to use the custom LLM features in a dockerized version of Draw.io?

I already tried
docker run -it --rm --name="draw" -p 8080:8080 -p 8443:8443 jgraph/drawio

adding above lines in Extras, Configuration tells me "Changes will take effect after a restart of the application."
After restarting the container I see the sparkle icon. But even the simplest prompt input causes the Output: Error: 0

Is my above 5 line JSON configuration generally enough or do I need more? Do I need to create a docker volume? If so how?

[alderg]

The AI feature works in Docker, but there are two things to configure:

1. Enable AI explicitly

enableAi defaults to true only on app.diagrams.net. In Docker (or desktop), you need to set it in your configuration:

{
  "enableAi": true,
  "gptApiKey": "sk-proj-..."
}

In Docker, pass this via the DRAWIO_CONFIG environment variable, or users can set it via Extras > Configuration.

2. Update the Content Security Policy

The Docker image sets a strict CSP that does not include the LLM API domains. Even with valid keys, the browser will silently block the requests. You need to override DRAWIO_CSP_HEADER to add the relevant API endpoint(s) to connect-src:

• OpenAI: https://api.openai.com
• Anthropic/Claude: https://api.anthropic.com
• Gemini: https://generativelanguage.googleapis.com
• Custom endpoint: whatever URL you set in gptUrl

Example docker-compose snippet:

environment:
  DRAWIO_CONFIG: '{"enableAi":true,"gptApiKey":"sk-proj-..."}'
  DRAWIO_CSP_HEADER: "default-src 'self'; script-src 'self' https://storage.googleapis.com https://apis.google.com https://docs.google.com https://code.jquery.com 'unsafe-inline'; connect-src 'self' https://api.openai.com https://*.dropboxapi.com https://api.trello.com https://api.github.com https://raw.githubusercontent.com https://*.googleapis.com https://*.googleusercontent.com https://graph.microsoft.com https://*.1drv.com https://*.sharepoint.com https://gitlab.com https://*.google.com https://fonts.gstatic.com https://fonts.googleapis.com; img-src * data:; media-src * data:; font-src * about:; style-src 'self' 'unsafe-inline' https://fonts.googleapis.com; frame-src 'self' https://*.google.com;"

The CSP issue is the main reason the sparkle button may appear but API calls fail silently.