Excalidraw 2.23.0-beta-x

Obsidian Excalidraw Beta releases
21

2.23.0-beta-6

Further, more invasive code quality changes to meet Obsidian guidelines.

At this point I will stop working on code quality. Use Excalidraw intensively the next two days, and hope to release 2.23.0 on Sunday. Hopefully nothing broke, but there can be hidden issues that are hard to spot.

If possible I’d love to hear some feedback about AI features in the mean time, so I can publish something solid.

22

You are just too fast :laughing: I didn’t get around to more testing yesterday but will take some time today and tomorrow!

23

Yeah, there is just too much noise around published reports that I fear undermine trust in the plugin, even though without any basis. I may actually push the update live today to try and address some of the biggest concerns.

24

2.23.0-beta-7

  • Implemented github signed automated release workflow
  • Updated plugin readme with information about codescanner findings
  • Upgraded the plugin to Node 22.
  • removed dotenv and fs-extra packages
  • replaced the unmaintained js-yaml with the yaml pacakge.
25

Hi Zsolt,

I played around a bit and didn’t come upon any strange behavior as far as general functionality is concerned. AI functionality is another topic but I’m obviously using a rather exotic setup. Below are a few more observations in case you’re interested – if not, I’m off work next week and will finally find a few hours to look under the hood and (hopefully) figure out what the issue is.

Setup: OpenAI-compatible (Obsidian AI Tools) + different models
  • Create a simple Excalidraw icon works reliably using models chatgpt-latest, claude-latest and gemini-latest
  • Same is true for Convert sketch to shapes and Challenge my thinking
  • Create Mindmap also works reliably with all three models (I consider this enough added benefit :smiley: )
  • All tasks involving image generation fail, independent of the chosen model
    • Typical error: “Uncaught (in promise) TypeError: Cannot read properties of null (reading ‘put’)”
    • Other times, errors do not even appear in the console even with verbose logging enabled (is this normal?)
    • With Generate an image from image and prompt the task still fails. However, in the console I see that at least the text generation has worked.
Setup: Local (Ollama) + different models
  • Not a single task successfully completed, yet. Still trying to figure out if this is related to wrong endpoints or the API key topic (key is not required and cannot be added manually, maybe Ollama doesn’t like being passed a dummy either?)
26

I ended up purchasing API keys in all 4 large models so I can test:

  • Anthropic/Sonnet
  • xAI/Grok4
  • OpenAI/GPT-5 & image-2
  • Google/Gemini-2.5 and 3

Each model worked as expected without any hick ups using the default configurations that come with the plugin out of the box. Image generation, mind map generation, etc. Keep in mind providers have their own limitations. Sonnet does not support image generation. Gemini and Grok do not support image masking. However, within these limitations everything functions correctly.

I do not use local LLMs, nor plan to use one in the foreseeable future… So that feature will remain untested/not-working, until someone comes along and submits a PR if required.

As for 3rd party gateway providers, I do not plan to support those for now. The issue is, that there are important differences between what each of the big 4 offer. Their APIs differ and not everything translates well to an OpenAI compatible endpoint. The code currently addresses each using their respective protocols.