MK2k

A lightweight HTTP proxy that translates OpenAI /v1/chat/completions requests to Ollama's native /api/chat API and back.

Ollama's OpenAI-compatible endpoint (/v1/chat/completions) has serious runtime configuration limitations.

Without custom model files, parameters such as context size cannot be centrally controlled. Requests through the OpenAI compatibility layer may run with Ollama defaults instead of the desired runtime settings.

This proxy uses Ollama's native /api/chat endpoint directly, where runtime parameters are explicitly supported.

One .env file configures all models centrally.

"CUDA on NVIDIA, MPS on Apple Silicon, CPU fallback" - any chance for some AMD love?

.werkkzeug1 was also my first though while scrolling through the article

The "4" clue forces the solution imho.

this needs to be an offline bootable usb version :)

visited expecting some bedroom DJing topics, stayed for actual agricultural machinery :)

The content analysis can be off. The following text has been flagged "negative":

> STOP asking Claude Code to design your apps. It’s great for logic, not design. Google Stitch 2.0 + Claude Code via MCP is the workflow that finally fixes the AI slop problem.

Vibe coded CRMs do exist. They are not open to the public though, as they are just used by the company that built them tailored to their needs is all.

Same with more complex systems: entire shop systems with payment integration, ERP and all – heavily supported by LLM code tools with a 3-10x productivity boost, just done by the CTO and no additional developers needed. They exist, the shop greets it’s customers and all you see is Vue and Tailwind as the tech stack where Shopify could’ve been. It’s now completely owned by the company selling things (they just don't sell the software).

front page, scroll down, there's Developers - Source Code

The first random blog post that I got served was this one:

https://nutcroft.com/blog/goodbye/

> [...] And now this blog goes too. So, goodbye!

Oh well

Nice tool! I tested it against imdb.com with its large graphql responses and after a few requests API Parrot crashed.

Try a Hackintosh, e.g. as a virtual machine on your Windows or Linux host.

That map idea is super cool, thanks for sharing!

TL;DW: Linus Åkesson basically live codes a step sequencer on the C64 in BASIC

Maybe off-topic, but: is "just closing without comment or discussion" an acceptable way of a maintainer to deal with pull requests?

Asking as someone who only occasionally contributed (or tried to) with a repository.

Examples: https://github.com/ben-n93/SQL-tips-and-tricks/pulls?q=is%3A...

Not sure what you mean by "media scanning", but if you assume that Media Hoarder isn't able to scan directories of your media, then I need to update the README - because it absolutely DOES this.

[Edit]: I read your post a second time, you use the term "media scanning" for two different things:

1/ Scanning in files

2/ Providing Metadata

Both are supported by Media Hoarder, none of the metadata you see in the screenshots and demos are manually provided by the user.

If you want to dig a bit deeper (coming from the README), you can see linked documents that go into detail for both topics:

https://github.com/theMK2k/Media-Hoarder?tab=readme-ov-file#...

You're welcome :)

I didn't plan to do direct comparisons, but it's not the first time someone asked about this, I might have to do it...

I might introduce an option for self-defining the color scheme. Care to head over to GitHub and create an issue on it and volunteer for testing?

> playing files

Thanks for pointing this out. Of course, Media Hoarder "plays files" - by launching the media player of your choice. I quickly updated the README :|

If "alongside Plex, Jellyfin" would be an intended use-case then I would've involved those names in the README or website.

Of course you can run Media Hoarder alongside them, it won't interfere, but that'd be about it.

First of all, I don't use any arr. From what I see the arr tools are used for obtaining media, less so presenting and providing a UI. Correct me if I'm wrong, but I assume that *arr tools and Media Hoarder might actually go well together.

Thanks, I wasn't aware that it might leave room for interpretation or confusion.

Still based on files, however, Media Hoarder doesn't care if you feed it 0-byte files. So yes, you can export your Excel file as such.

Actually, I do use 0-byte files in my tests and demos. They are artificially created in the following way: https://github.com/theMK2k/media-hoarder-testset-generator

My pacifist nature lead the choice and I probably saw that license on some other random projects iirc.

this made me chuckle, yes, it's not gonna happen :D

Yes, someone (you?) can create a "Kodi .nfo" exporter from the SQLite database.

You're right and I am and was hesitant to put Media Hoarder directly against established solutions on the front page.

I created a blog post about this when v1.0.0 was released (maybe a bit too hidden): https://media.hoarder.software/blog/why-media-hoarder

Feel free to open up an issue on GitHub, maybe we introduce an option on how heatmaps are shown :)

It's more Kodi-like (no client-server) necessary.

The only "AI" in it is that you can run a ChatGPT session which creates a filter (e.g. "give me the top 10 oscar winning movies of the 90s"). It's a nice feature but not the heart of Media Hoarder. -> https://www.youtube.com/watch?v=Pi2vfPpOEGQ

Trying to submit something, you are bound to a fixed list of categories and alternatives.

If what you want to submit doesn't fit the fixed values in there, you are blocked.