Looks like something minecraft-related (I’ve never played it, i just know the blocky look).

Website doesn’t display well on phones.

This project is licensed under the Open WebUI License, a revised BSD-3-Clause license. You receive all the same rights as the classic BSD-3 license: you can use, modify, and distribute the software, including in proprietary and commercial products, with minimal restrictions. The only additional requirement is to preserve the “Open WebUI” branding, as detailed in the LICENSE file. For full terms, see the LICENSE document.

The license seems to me to say that you can use, modify, redistribute Open-WebUI as you want, but you can’t change the branding unless your instance has <= 50 users (or a couple other conditions are met).

I’m not familiar with Caddy at all - I use Traefik for a reverse proxy, and my knowledge there isn’t huge either. But I think that your reverse proxy terminates TLS (HTTPS) from the world and then forwards traffic to the appropriate service on your local network using HTTP by default - but if your local service can handle TLS, I think you can configure your reverse proxy to forward the traffic to it using TLS.

IT-Tools is kind of fun: a web page full of common tools, converters, references, cheat sheets, etc.

A few of you mentioned the “indent” - I never even saw it until I zoomed way in on the photo I took after you mentioned it. I can’t even see it in person. My vision isn’t great, but I think that suggests even more strongly a bad design.

I agree, if you have POTS then you can send a fax if the printer has a modem installed. I don’t happen to have a POTS landline.

As others said, Searxng is a search proxy. Any given instance can be configured to return results from vary many sources. Why would the results vary with repeated identical search? Not sure, but perhaps a bit of randomness is in the algorithm of which sites yo return results from, or which order to display them in. Also possible: latency may vary and so for each search a different result set is retrieved first.

Thanks for that explanation, @[email protected]!

Googled earned about $325B in 2024. $30M is about .009% of that. If you earn $50K per year, it would be like you had to pay $4.50 for committing a crime.

It’s been a few weeks, @[email protected] … Did you release anything? Any hints as to what may be coming?

I also like/use CryptPad. FYI, it’s self-hostable.

Cryptpad is also self-hostable, if that’s your bag.

Wow that’s a lot to process! Thanks for the explanations, #[email protected] !

You suggested a 20b model but OP said they have 8G VRAM (and 32G RAM). I thought a [very] rough approximation of VRAM size needed for a model was 1:1 with billion parameters (so 20G VRAM in this example). It seems I’m wrong. What’s a better approximation or heuristic to use?

Here’s the Readme from the commit when he discontinued Tesseract.

Here’s a post where the developer explains his reasons.

I use Tesseract, which is fantastic. Be wary, though: the developer recently put the project in maintenance mode and swore off the fediverse, then i guess changed his mind and continues to develop it - it’s certainly possible it could happen again.

Let’s find out. One… two-hoo… three. CRUNCH. Three.

Thanks for the explanations, @[email protected].

I introduced myself to tools/functions in Open-WebUI: there is a community library of tools/functions and you can essentially copy/paste them into Open-WebUI and then they’re available there. If I understand correctly, tools add functionality to models and functions add functionality to Open-WebUI.

I’m still confused, though: if I add a web search tool to Open-WebUI (some do exist), I would guess that it could only ever add functionality to the model if I access the model through Open-WebUI - but HA is accessing the model directly through Ollama.

Thanks for the reply and the assistance, @[email protected].

I’m not using any add-ons in Home Assistant for this integration, just the Ollama integration (but I suspect that’s what you meant).

Its docs say that it doesn’t integrate with sentence triggers, so I think this suggestion won’t work. :(

This is interesting. Thanks for sharing your test results.

Do I understand correctly that if your computer shares a fingerprint with fewer computers, it’s more distinguishable/identifiable?