Stay updated with the latest posts from Thomas Wilde Tech!https://thomaswildetech.com/ en Sun, 11 Oct 2026 05:59:52 -0000 Sun, 11 Oct 2026 05:59:52 -0000 1440 MkDocs RSS plugin - v1.19.0 None https://thomaswildetech.com/ Thomas Wilde AI AdGuard DeepSeek Hermes OpenClaw Self-Hosting local-ai opencode pi privacy I cloned five AI agent harnesses — opencode, pi, DeepSeek Harness, Hermes, OpenClaw — grepped every one for telemetry endpoints, then cross-checked against 24 hours of whole-LAN DNS logs. Here's exactly what phones home, and what doesn't. https://thomaswildetech.com/blog/2026/10/10/do-ai-coding-agents-phone-home-i-grepped-five-of-them/ Sat, 10 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/10/do-ai-coding-agents-phone-home-i-grepped-five-of-them/ Thomas Wilde AI Android DeepSeek Hermes Self-Hosting local-ai opencode pi qwen-code opencode, pi, DeepSeek Harness, and qwen-code compared on the features that matter for a self-hosted setup: web server modes, multi-turn goal prompts, messaging integrations, and the open-source Android clients that drive them from the couch. https://thomaswildetech.com/blog/2026/10/10/four-open-source-coding-agents-serve-modes-android-clients-and-chat-integrations/ Sat, 10 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/10/four-open-source-coding-agents-serve-modes-android-clients-and-chat-integrations/ Thomas Wilde AI Hermes Mattermost OpenRouter Self-Hosting Signal local-ai privacy How my AI agents live inside my own messaging: self-hosted Mattermost with a bot account and custom plugins (the !-prefix trick and in-screen button menus), Signal through a dedicated Google Voice number, aux models with thinking turned off, and what I insist on before any OpenRouter fallback. https://thomaswildetech.com/blog/2026/10/10/give-your-ai-agent-a-private-messaging-home-signal-and-self-hosted-mattermost/ Sat, 10 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/10/give-your-ai-agent-a-private-messaging-home-signal-and-self-hosted-mattermost/ Thomas Wilde 5090 AI MoE Qwen Software Development Strata local-ai How I actually run Strata day to day: IQ3_S on a 5090 with 128 GB DDR5, 260K context, Hermes agents and Opencode on top, conversation parking for concurrency, and elastic VRAM resize so agents can queue ComfyUI jobs without losing their brain. https://thomaswildetech.com/blog/2026/10/08/running-strata-on-a-5090-the-setup-i-actually-use/ Thu, 08 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/08/running-strata-on-a-5090-the-setup-i-actually-use/ Thomas Wilde Fedora Hermes KVM Self Hosted backups libvirt My Fedora KVM host ran Hermes agent VMs with nothing but manual virt-manager snapshots. Here's the script and systemd timer that replaced that habit, and the pile of disk it cleaned up. https://thomaswildetech.com/blog/2026/10/08/the-manual-snapshot-habit-that-ate-400-gb/ Thu, 08 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/08/the-manual-snapshot-habit-that-ate-400-gb/ Thomas Wilde Projects Self-Hosting homelab umbrel umbrelOS virtualization umbrelOS 2.0 in a VM: what's new, whether nested VMs (Machines) work, and how it compares to a DIY Proxmox + Docker stack https://thomaswildetech.com/blog/2026/10/04/umbrelos-20-whats-new-and-can-it-run-vms-inside-a-vm/ Sun, 04 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/04/umbrelos-20-whats-new-and-can-it-run-vms-inside-a-vm/ Thomas Wilde GPU Proxmox RAM Self Hosted I doubled the Powerspec box to 128 GB, sized a 114 GB VM, and the host OOM'd anyway. The kernel was holding 8.5 GB that every memory counter called free. https://thomaswildetech.com/blog/2026/10/02/the-85-gb-that-showed-up-as-free/ Fri, 02 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/02/the-85-gb-that-showed-up-as-free/ Thomas Wilde AI GPU HTTP Projects llama-slot-proxy Why the proxy answers 429 instead of 503 when the GPU is busy with a media job https://thomaswildetech.com/blog/2026/10/01/why-i-send-429-instead-of-503/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/why-i-send-429-instead-of-503/ Thomas Wilde AI Docker GPU Projects A container runs whatever it was last created with, not what's in the compose file https://thomaswildetech.com/blog/2026/10/01/a-container-runs-what-it-was-last-created-with/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/a-container-runs-what-it-was-last-created-with/ Thomas Wilde AI AI GPU benchmarking Decode tokens per second is a real number and it's the wrong one to judge a model by https://thomaswildetech.com/blog/2026/10/01/the-number-that-lies-to-you-decode-tokens-per-second/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-number-that-lies-to-you-decode-tokens-per-second/ Thomas Wilde AI AI Projects deploy llama-slot-proxy Deploying a proxy that serves the agent doing the deploying https://thomaswildetech.com/blog/2026/10/01/deploying-a-proxy-to-itself/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/deploying-a-proxy-to-itself/ Thomas Wilde AI GPU cache llama.cpp A llama.cpp bug that isn't in upstream, and the 0% cache after a slot restore it caused https://thomaswildetech.com/blog/2026/10/01/the-bug-that-wasnt-in-upstream/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-bug-that-wasnt-in-upstream/ Thomas Wilde AI Docker GPU Projects llama-slot-proxy What the slot proxy is for, and why multi-agent local inference needs a front layer https://thomaswildetech.com/blog/2026/10/01/a-front-layer-for-local-inference/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/a-front-layer-for-local-inference/ Thomas Wilde AI AI GPU Proxmox Self Hosted Proxmox hot-shrank a VM from 48GB to 16GB on the same boot and the engine refused to start https://thomaswildetech.com/blog/2026/10/01/the-proxmox-balloon-ate-my-model/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-proxmox-balloon-ate-my-model/ Thomas Wilde AI AI benchmarking reasoning Three requests burned 13 minutes of thinking and produced no answer https://thomaswildetech.com/blog/2026/10/01/runaway-thinking/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/runaway-thinking/ Thomas Wilde Docker Mealie Self Hosted gotify ntfy Gotify kept pinging 401s that looked like an app was erroring. It was a log matcher, and the real cause was a kiosk session expiring. https://thomaswildetech.com/blog/2026/10/01/the-401-spam-that-wasnt-ha/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-401-spam-that-wasnt-ha/ Thomas Wilde GPU Proxmox RAM Self Hosted I sized a 56 GB VM on a 64 GB host and the OOM killer took kvm twice. Here's what I had wrong and what the numbers actually said. https://thomaswildetech.com/blog/2026/10/01/the-56-gb-that-wasnt-there/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-56-gb-that-wasnt-there/ Thomas Wilde Monitoring PBS Proxmox Self Hosted A storage pool flapped to inactive during an auth refresh and the watchdog called it CRITICAL. That false alarm became the debounce design. https://thomaswildetech.com/blog/2026/10/01/the-false-alarm-that-became-the-watchdog/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-false-alarm-that-became-the-watchdog/ Thomas Wilde Debugging Home Assistant Self Hosted Voice A missing byte in my own test client looked like a broken voice pipeline for weeks. The satellite was never broken, only the harness was. https://thomaswildetech.com/blog/2026/10/01/the-one-byte-bug/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-one-byte-bug/ Thomas Wilde PBS Proxmox Self Hosted virtiofs Virtiofs ENFILE errors failed vzdump about half the time on the external drive. The fix was not a better mount, it was a different kind of passthrough. https://thomaswildetech.com/blog/2026/10/01/the-stale-mount-that-ate-half-the-backup-windows/ Thu, 01 Oct 2026 00:00:00 +0000 Thomas Wilde Techhttps://thomaswildetech.com/blog/2026/10/01/the-stale-mount-that-ate-half-the-backup-windows/