Been running Open WebUI on a little 2 GB / 1 vCPU VPS, API models only, no local LLM. Sharing because the RAM story wasn't what I expected.

Just chatting? Totally fine. It mostly sits there forwarding requests and streaming replies back.

The thing that bit me was document upload. The default image does embeddings locally (a small MiniLM model), so the moment someone drops a stack of PDFs into a knowledge base the CPU pins and memory climbs. On 2 GB that was where I started seeing the container get OOM-killed.

What helped:

• pointing embeddings at an API instead of computing them locally
• just going to 4 GB once uploads became a daily thing
• setting WEBUI_SECRET_KEY to a fixed value. before that, every redeploy logged everybody out and I thought I'd broken something

Still on SQLite. Has anyone squeezed Postgres onto the same tiny box, or is that asking for trouble?

Omada platform was found to be vulnerable in several ways

Apparently Gemma 4 (by deepmind) are very energy-efficient for their capabilities.

The post itself seems to recommend a company (neuralwatt.com) but it's unrelated to project wagtail (an open source CMS made with django) so I consider it legit, they're just vouching for this new approach that shows users their ecological footprint basically.

(Of course, running these at a homelab is possible. hence the relevance of this info for the community)

cross-posted from: https://programming.dev/post/40013730

Cross-posted from: https://programming.dev/post/33674513

Sascha Koenig looks to have some quality deep guides around NixOS!

This is a "nextcloud" or "owncloud" alternative. Its origin was as a fork to owncloud infinite scale (ocis), so it's shares the underlying feature set (written in Go instead of PHP, being lenient on resources and usable on a raspberry pi 4B.)

The latest release (2.0, see link) leverages PosixFS so instead of having to manage a database, you can rely on snapshots (on the filesytem) for all your backups.