Skip to content

Instantly share code, notes, and snippets.

View AlexsJones's full-sized avatar

Alex Jones AlexsJones

View GitHub Profile
@AlexsJones
AlexsJones / README.md
Last active August 27, 2026 13:23
Rootless install of the Grok Bot .deb on Fedora — no alien, no dpkg, no root

Installing the Grok Bot .deb on Fedora, rootless

The newest native Linux build of Grok Bot was distributed as a Debian package. On Fedora there is no dpkg, and the usual alien route wants rpm-build, a Perl stack, and root. None of that is necessary: the package is an Electron app that lives entirely under /opt, so it can be unpacked straight into ~/.local and run as your own user.

install-grok-bot.sh does that in one shot. Updated for Grok_Bot_0.24.0.deb on Fedora 44.

@AlexsJones
AlexsJones / # nouscell - 2026-08-02_13-40-50.txt
Created August 2, 2026 12:41
nouscell (sympozium-ai/tap/nouscell) on Fedora Linux 44 (KDE Plasma Desktop Edition) - Homebrew build logs
Homebrew build logs for sympozium-ai/tap/nouscell on Fedora Linux 44 (KDE Plasma Desktop Edition)
Build date: 2026-08-02 13:40:50
@AlexsJones
AlexsJones / deepseek-v4-flash-strix-halo.md
Created July 5, 2026 18:49
Running DeepSeek-V4-Flash (284B MoE) on a 64GB Strix Halo via SSD expert-streaming

Running DeepSeek-V4-Flash (284B MoE) on a 64 GB Strix Halo by streaming experts from SSD

A 284B-parameter mixture-of-experts model, quantized to 2-bit, is 81 GB — bigger than the 62 GiB of RAM on an AMD Strix Halo (Ryzen AI Max+ 395). It still runs at ~1.9 tok/s, coherent, because only 6 of 256 experts fire per token: you mmap the file, keep the ~13 GB dense/attention core hot in page cache, and stream the cold 2-bit experts off the NVMe SSD on demand.

CPU-only, ~1.9 tok/s decode / ~6 tok/s prefill. Slow, but it runs — a usable offline reasoning tool.

Hardware

@AlexsJones
AlexsJones / local-llm-setup-guide.md
Last active May 5, 2026 07:16
Local LLM coding setup: Framework Desktop + Tailscale + llama-server + OpenCode

Local LLM Setup: Framework Desktop + Tailscale + OpenCode + Hermes Agent

Run a local LLM inference server on a beefy desktop and use it as a coding assistant from anywhere on your tailnet.

Hardware

Role Machine Specs
Server Framework Desktop AMD Ryzen AI MAX+ 395 w/ Radeon 8060S, 128 GB unified RAM, 8 TB HDD, Fedora
Client MacBook Pro 14" Apple M5 Max, 128 GB RAM
@AlexsJones
AlexsJones / run.sh
Created March 27, 2025 15:49
try k8sgpt with kind
#!/bin/bash
# Default values, can be overridden with environment variables or command-line args
KIND_CLUSTER_NAME="${KIND_CLUSTER_NAME:-k8sgpt-cluster}"
HELM_RELEASE_NAME="${HELM_RELEASE_NAME:-k8sgpt}"
NAMESPACE="${NAMESPACE:-k8sgpt}"
OPENAI_TOKEN="${OPENAI_TOKEN:-}"
# Check for OpenAI token
if [[ -z "$OPENAI_TOKEN" ]]; then
@AlexsJones
AlexsJones / workspaces-linux-client-rpm.md
Created August 7, 2024 16:03 — forked from nhira/workspaces-linux-client-rpm.md
How to install Amazon Workspaces Linux Client for Fedora

Amazon WorkSpaces Linux client - RPM installation

Overview

The good news is that there is an Amazon WorkSpaces Linux client. If you happen to use Ubuntu, you can use the official instructions from the documentation. But what if you're on an rpm-friendly distribution like Fedora or CentOS?

alien to the rescue.

How to

You can use these steps to build a PCoIP client (not the 2023 WSP client) RPM from the original .deb package.

@AlexsJones
AlexsJones / README.md
Created January 10, 2024 17:27
Fix neovim on maca

Install .rustup directory in home, otherwise there are a bunch of rust-analyzer issues

@AlexsJones
AlexsJones / cp-0
Created March 6, 2023 08:34
DQLite debugging
ubuntu@raspbernetes-0:~/dqlite-issue-repro$ cat /var/snap/microk8s/current/var/kubernetes/backend/cluster.yaml
- Address: 192.168.88.239:19001
ID: 3297041220608546238
Role: 0
- Address: 192.168.88.249:19001
ID: 10829805846545420511
Role: 0
- Address: 192.168.88.246:19001
ID: 8819719984785607608
Role: 0

hardware

  • pi4 x3 control-plane nodes
  • pi4 x1 worker nodes
  • Attached disks on /dev/sda on each node

Install

I don't think I enabled coreDNS by default

You must run these commands on all nodes, including the workers ( especially cgroup enablement for containerd )

sudo swapoff -a
sudo mkdir -p /etc/apt/keyrings
curl -fsSL https://download.docker.com/linux/ubuntu/gpg | sudo gpg --dearmor -o /etc/apt/keyrings/docker.gpg
echo "deb [arch=$(dpkg --print-architecture) signed-by=/etc/apt/keyrings/docker.gpg] https://download.docker.com/linux/ubuntu $(lsb_release -cs) stable" | sudo tee /etc/apt/sources.list.d/docker.list > /dev/null