SumGuy's Ramblings
The art of wasting time.
Docker, self-hosting, AI/LLM, Linux, and DevOps — explained by someone who learned the hard way. No fluff, no enterprise jargon, just practical stuff that actually works on real hardware.
Recent Posts
-
Batch Inference: One Model, 100 Users
Loading model weights, not math, is your GPU bottleneck. Continuous batching with llama-server serves many concurrent requests for close to the cost of one.
12 min read -
Stop Handing CI Jobs Root on Your Box
Self-hosted GitHub Actions, Forgejo, and GitLab runners can hand a forked pull request root on your box. Here is the isolation ladder that actually stops it.
13 min read -
OPNsense Multi-WAN Failover Guide
Configure OPNsense gateway groups for WAN failover and load balancing, then layer policy-based routing so specific hosts or VLANs use a chosen WAN link.
14 min read -
Cluster API for Homelab k3s
Cluster API turns k3s cluster lifecycle into YAML: create, scale, and upgrade whole clusters as code with a management cluster, no more one off scripts.
11 min read -
RabbitMQ vs NATS: Pick a Message Bus
RabbitMQ vs NATS for a self-hosted app: routing, delivery guarantees, Docker Compose setups, Python code, and which broker to run for background jobs.
14 min read -
Meilisearch vs Typesense for Self-Hosters
Meilisearch and Typesense both promise instant search without Elasticsearch's weight. I ran both in Docker and picked a winner per self-hosted use case.
13 min read -
The Update That Broke Confidence
A 2026-09-22 patch left a self-hosted model's weights untouched but made it more confident on nonsense input. Six systems compared, real numbers only.
· Updated:18 min read -
CrowdSec Across a Home Lab: One Brain
One central CrowdSec LAPI, remote log-parsing agents, and bouncers on every box, so an SSH brute-force on one server gets blocked everywhere else too.
12 min read -
Docker MCP Toolkit: Trust the Sandbox?
Docker MCP Toolkit sandboxes AI agent tools in containers. Here's exactly what that isolation stops, where it silently breaks, and how to configure it safely.
12 min read -
Jev vs Von vs SemIf: Real Numbers
Jev, Von, and SemIf scored head to head on real commit history: a frozen 4B general model beat a purpose-built 395M decision model on every task tested.
14 min read -
NetAlertX MCP: One Token, Twelve Tools
NetAlertX v26.9.0 ships a built-in MCP server with 12 tools. Learn how to wire it into Claude Code, what it exposes, and when to use Home Assistant instead.
13 min read -
Skeleton Key MCP vs One Token Per Box
Skeleton Key MCP swaps four narrow homelab tokens for one Vaultwarden vault and one OAuth consent flow. Here is what that buys you, and what it risks.
14 min read
