https://odysee.com/@HyperVegan:2/20261115-HyperVegan-LibreSoftwareMaxxing:08cf2da8ca


Let’s talk about free and open-source software!

In the last few months, I’ve changed operating systems, desktop environments, and service managers.

I’ve also made some pretty important changes to my video transcoding command and done quite a deep dive into learning local AI, how to run it properly on my hardware, and how to get it to actually provide productive and useful output.

Timestamps

00:00 Intro & Jokes

00:20 Problem Statement

01:28 i3 Window Manager

02:30 Trying Devuan (from Debian)

03:39 Runit Init System (from SystemD)

05:02 Moving to Artix (from Devuan)

05:41 Disadvantages of Arch (Run your updates!!)

06:53 Timeline of Operating Systems

07:23 Why I Say “Libre” and “FLOSS”

07:50 Advantages of Artix / Arch

08:36 Advantages of Provisioning Scripts

09:35 Joy of Linux

ffmpeg

10:10 Video Encoding Changes

10:40 Advantages of AV1

12:17 pv-test Bash Function

13:30 Video Codec Presets (Speeds)

Local AI

16:10 Intro – Why I Started

17:02 Tech Stack (Ollama, Open WebUI, Open Terminal)

19:05 Docker, AGENTS.md

19:49 Android Versioning Rant

21:08 Agents + Existing Projects

21:19 Performance, Run Parameters, Models

22:20 ollama-test Bash Function

23:37 swap-stopper Runit Script

24:05 Which Models For What

25:40 Speed / Tokens per Second!

27:02 What is an LLM?

27:32 Agents + New Projects

29:04 AI as a Junior Engineer

30:00 AI Botching Infinite Timer

30:28 Google’s Planned Obsolescence (Android)

31:37 Use Cases – Running on Solar!

31:48 Prompts & Safeguards

32:23 Docker & runit

33:37 Rogue Agent Story

34:34 AI as a Junior Engineer #2

34:55 Other People’s Opinions

35:59 Future Plans

37:43 AI Provisioning Files

38:50 Wrap Up

Using Runit (Devuan vs Artix)

The biggest disadvantage of using runit on Devuan was that they did not support virtualization on it, which is critical for me.

Android Virtual Devices (AVDs) would not boot and I don’t think Docker was working either. These work fine on Artix with runit, so not sure what their problem is.

Video Commands

Below are the ffmpeg commands that my process video (pv) functions call. It’s not the entirety of each function, it’s simply the portion that transcodes my edited video into the upload video.

pv is the original that I’ve been using for years to create H.264, AAC, MP4 videos. pv3 is the most recent version that I’m using to create the AV1, Opus, WebM version.

pv2 was a failed experiment to make the H264 videos more efficient with extra passes and a slower preset, but it actually ended up making larger files.

The full scripts can be found in my env-shared project which is in the Links section below.

Horizontal

pv

ffmpeg -nostdin -hide_banner -loglevel level+warning -i edited-file.mp4 -filter:v scale=-1:720 -b:v 2000k -minrate 0 -maxrate 2000k -bufsize 2000k -b:a 192k -filter:a dynaudnorm=f=33:g=65:p=0.66:m=33.3 -vcodec libx264 -movflags +faststart upload-file.mp4

pv3

ffmpeg -nostdin -hide_banner -loglevel level+warning -i edited-file.mp4 -c:v libsvtav1 -preset 4 -filter:v scale=-1:720 -svtav1-params rc=1:tbr=1200k:overshoot-pct=100:undershoot-pct=100 -c:a libopus -filter:a dynaudnorm=f=33:g=65:p=0.66:m=33.3 -b:a 128k -movflags +faststart upload-file.mp4

Vertical

pvs

ffmpeg -nostdin -hide_banner -loglevel level+warning -i edited-file.mp4 -filter:v scale=720:-1 -b:v 2000k -minrate 0 -maxrate 2000k -bufsize 2000k -b:a 192k -filter:a dynaudnorm=f=33:g=65:p=0.66:m=33.3 -vcodec libx264 -movflags +faststart upload-file.mp4

pv3s

ffmpeg -nostdin -hide_banner -loglevel level+warning -i edited-file.mp4 -c:v libsvtav1 -preset 4 -filter:v scale=720:-1 -svtav1-params rc=1:tbr=1200k:overshoot-pct=100:undershoot-pct=100 -c:a libopus -filter:a dynaudnorm=f=33:g=65:p=0.66:m=33.3 -b:a 128k -movflags +faststart upload-file.mp4

Local AI

My preferred setup has been with ollama as the backend, and then Open WebUI and Open Terminal running in separate Docker containers.

I have also tried experimenting with ZeroClaw, ZeroCode, Qwen Code (mostly using the serve command), and the Visual Studio Code / Codium plugins Continue and Cline.

So far my interest has only been with QWEN and Mistral models since they seem the most libre. Models like llama from Meta do not interest me since I don’t trust the companies even if they say they’re “community” projects.

Laptop Specs

Here are some details on my laptop’s hardware specifications.

$ fastfetch
                   '                       user@system
                  'o'                      ---------------
                 'ooo'                     OS: Artix Linux x86_64
                'ooxoo'                    Host: NV4xPZ (v2.1)
               'ooxxxoo'                   Kernel: Linux 7.2.15-artix1-1
              'oookkxxoo'                  Uptime: 55 mins
             'oiioxkkxxoo'                 Packages: 42 (flatpak), 1513 (pacman)
            ':;:iiiioxxxoo'                Shell: bash 5.3.20
               `'.;::ioxxoo'               Display (eDP-1): 1920x1080 in 14", 60 Hz [Built-in]
          '-.      `':;jiooo'              Window Manager: i3 4.25.1 (X11)
         'oooio-..     `'i:io'             Theme: Adwaita-dark [GTK2/3/4]
        'ooooxxxxoio:,.   `'-;'            Icons: Adwaita [GTK2/3/4]
       'ooooxxxxxkkxoooIi:-.  `'           Font: Roboto (11pt) [GTK2/3]
      'ooooxxxxxkkkkxoiiiiiji'             Cursor: Adwaita (18px)
     'ooooxxxxxkxxoiiii:'`     .i'         Terminal: kitty 0.50.0
    'ooooxxxxxoi:::'`       .;ioxo'        Terminal Font: DejaVuSansMono (11pt)
   'ooooxooi::'`         .:iiixkxxo'       CPU: 12th Gen Intel(R) Core(TM) i7-1260P (8+8) @ 4.70 GHz
  'ooooi:'`                `'';ioxxo'      GPU: Intel Iris Xe Graphics @ 1.40 GHz [Integrated]
 'i:'`                          '':io'     Memory: 5.57 GiB / 62.63 GiB (9%)
'`                                   `'    Swap: 999.34 MiB / 68.89 GiB (1%)
                                           Disk (/): 582.75 GiB / 847.71 GiB (69%) - ext4
                                           Local IP (wg0-mullvad): 10.162.88.249/32
                                           Battery (BAT): 100% [AC Connected]
                                           Locale: en_US.UTF-8

ollama-benchmark Results

Here are the results of running a test against a few QWEN and Mistral models.

$ QUIET=1 ollama-test "32000 128000 264000" "4" "0" "qwen3.8:27b qwen3.6:35b qwen3.5:35b qwen3.5:4b devstral-small-2:24b ministral-3:3b"
| Model                     |    HDD |  Context |    RAM | Threads | iGPU | Processor | Speed (t/s) | Time (s) |
| :------------------------ | -----: | -------: | -----: | ------: | ---: | --------: | ----------: | -------: |
| qwen3.8:27b               |   17GB |    32000 |   18GB |       4 |    0 |   100%CPU |        2.81 |       28 |
| qwen3.8:27b               |   17GB |   128000 |   19GB |       4 |    0 |   100%CPU |        2.71 |       37 |
| qwen3.8:27b               |   17GB |   262144 |   20GB |       4 |    0 |   100%CPU |        2.35 |       33 |
| qwen3.6:35b               |   22GB |    32000 |   22GB |       4 |    0 |   100%CPU |       12.28 |       30 |
| qwen3.6:35b               |   22GB |   128000 |   23GB |       4 |    0 |   100%CPU |       13.66 |       27 |
| qwen3.6:35b               |   22GB |   262144 |   24GB |       4 |    0 |   100%CPU |       13.72 |       25 |
| qwen3.5:35b               |   22GB |    32000 |   22GB |       4 |    0 |   100%CPU |       13.76 |       62 |
| qwen3.5:35b               |   22GB |   128000 |   24GB |       4 |    0 |   100%CPU |       13.61 |       63 |
| qwen3.5:35b               |   22GB |   262144 |   28GB |       4 |    0 |   100%CPU |       13.60 |       40 |
| qwen3.5:4b                |  3.3GB |    32000 |  4.2GB |       4 |    0 |   100%CPU |       10.60 |       53 |
| qwen3.5:4b                |  3.3GB |   128000 |  8.0GB |       4 |    0 |   100%CPU |       10.20 |       41 |
| qwen3.5:4b                |  3.3GB |   262144 |   12GB |       4 |    0 |   100%CPU |       11.00 |       27 |
| devstral-small-2:24b      |   15GB |    32000 |   20GB |       4 |    0 |   100%CPU |        2.36 |      108 |
| devstral-small-2:24b      |   15GB |   128000 |   36GB |       4 |    0 |   100%CPU |        2.59 |      104 |
| devstral-small-2:24b      |   15GB |   264000 |   59GB |       4 |    0 |   100%CPU |        2.52 |      103 |
| ministral-3:3b            |  3.0GB |    32000 |  5.9GB |       4 |    0 |   100%CPU |       13.08 |       17 |
| ministral-3:3b            |  3.0GB |   128000 |   16GB |       4 |    0 |   100%CPU |       15.35 |       16 |
| ministral-3:3b            |  3.0GB |   262144 |   31GB |       4 |    0 |   100%CPU |       15.33 |       17 |

 Done!! :D

Speed Champion: ministral-3:3b at 4-128000 with a tokens/second of 15.34595379042985!
 Time Champion: ministral-3:3b at 4-128000 with a time of 16!

env-ansible – Desktop Provisioning

The main script that my desktop systems use which keep them consistent with each other if I make modifications to one, as well as allows me to distro-hop with ease.

git.hyperling.com/me/env-ansible

The ai folder is here:

tasks/workstation/linux/software/ai/

Docker network watcher is in here:

docker.yml

env-shared – Shared by Desktop and Mobile

Bash commands, functions, and aliases that I use across both the desktop and mobile, which do not have any special requirements or dependencies which require them to be either in my Ansible or Termux environment projects.

git.hyperling.com/me/env-shared

Some commands were made cross-platform by checking whether the SD card folder exists to know whether or not it’s being run on Termux (/sdcard). Most of them “just work” because all they are is Bash code.

env-termux – Android Provisioning

This one was not mentioned in the video, but is basically the mobile equivalent of my env-ansible project.

git.hyperling.com/me/env-termux

Contains any Turmux and Android specific setup, as well as mobile versions of things that have a desktop-only version.


Filmed and edited 2026-11-15. Written 2026-11-17.

blog.hyperling.com/libre-software-maxxing