Fix for incorrect local file paths in timeline events preventing image display

[rorygallagher2024]

Hi,
I've been experiencing an issue with the integration that I thought I'd share.

This is potentially a bug or feature request, so figure I'd share this as a discussion.

So for context I'm running Home Assistant Core 2025.5 (running bare metal on Asahi Linux) and I suspect the fact I'm running core is probably the key factor as to why I'm experiencing the issue. I'm currently running the latest versions of the LLM Vision Integration (1.4.3), and LLMVision timeline cards.

I have an automation setup to analyse my doorbell images on doorbell press or motion detection events.

Currently, when the expose_images: true option is used with the llmvision.image_analyzer service, the key_frame value stored in the events.db for the LLM Vision Timeline contains the full local file path to the saved image (e.g. /home/homeassistant2/.homeassistant/www/llmvision/your_image.jpg).

This prevents the images from being displayed correctly in the Home Assistant frontend (e.g., in the LLM Vision Timeline card) because the browser or app cannot access these local server paths directly via HTTP. The expected behavior is that the key_frame should contain a web-accessible URL for the image.

So, for example, my web browser tries to get the images like this: http://MYIPADDRESS:8123/home/homeassistant2/.homeassistant/www/llmvision/c776ed9a-0.jpg and a HTTP 404 is returned.

Proposed Solution:

I have been able to implement a solution by modifying the _expose_image function in the integration media_handlers.py file to generate and store the web-accessible /local/... path for the image instead of the full local file path.

The following code change in the _expose_image function (in the media_handlers.py file) resolves this issue:

Diff

--- a/media_handlers.py
+++ b/media_handlers.py
@@ -XXX,X +XXX,X@@
     filename = self.hass.config.path(
         f"www/{DOMAIN}/{uid}-{frame_name}.jpg")
-    self.key_frame = filename
+    self.key_frame = f"/local/{DOMAIN}/{uid}-{frame_name}.jpg"
    if image_data is None and frame_path is not None:
        # open image in hass.loop
        with await self.hass.loop.run_in_executor(None, Image.open, frame_path) as image:

By changing self.key_frame = filename to self.key_frame = f"/local/{DOMAIN}/{uid}-{frame_name}.jpg", the timeline events now correctly reference the image via the /local/ alias, allowing them to be displayed in the frontend for users experiencing this issue.

As a result, my frontend is able to grab the images with this URL: http://IPADDRESS:8123/local/llmvision/2efa0499-0.jpg

Can I suggest incorporating this change into the integration's code to ensure that the timeline correctly displays image snapshots for all users, especially those on setups like Home Assistant Core where the local pathing seems to be an issue?

Note: Whilst this fix works for me and will resolve the issue for setups where local paths are not correctly interpreted, I'm not 100% clear that this wouldn't negatively impact users whose image display is currently working across the different Home Assistant installation methods. Someone with experience may be able to say for sure!

Thanks!

[valentinfrlch]

Thanks for the suggestion!
You are right, the current implementation is far from perfect. The reason it doesn't work in your case, is because your core installation has a different folder structure. The card would replace the /config/www with /local, but fails since that isn't part of your file path. This is of course suboptimal.
However, I am currently trying to get rid of the www folder entirely (because it is exposed to the internet so it could theoretically be accessed by anyone...). A better alternative would be the /media folder as it would still expose its content but requires authentication.
Once I figure out how to generate an auth token from the frontend, I'll implement that change (and make sure it's compatible with the other folder structures - such as core - as well).

[rorygallagher2024]

Nice, thank you. I really appreciate the integration!