I build software and run the servers it lives on. Here I write down what AI tools actually do when I run them.
Senior full-stack developer and DevOps engineer. Reviews of AI tools, agents and self-hosted software, each with a verdict up top, and a monthly table of what model tokens cost.
- Work
- Full-stack web, CI/CD, containers, cloud and on-prem infrastructure
- Writes
- Hands-on reviews, the AI token market, homelab notes
- Contact
- [email protected]
Latest tables
All posts- 14 Market AI token market, November 2026Verdict: output tokens still drive the bill; the small tiers now cover most agent sub-tasks.
- 13 Review Ollama on one homelab GPU: three weeks as my local model serverVerdict: right for one or two people on one box; move to vLLM when a team shares the GPU.
- 12 Homelab Moving my Git hosting to ForgejoVerdict: quiet and light on memory; the Actions runner needs babysitting.
- 11 Review A coding agent on a real backlog for two weeksVerdict: good at the dull tickets; review every diff as if a stranger wrote it.
- 10 Market AI token market, October 2026Verdict: cached input became the price that matters for agents.
Key
Recommended
Use with caveats
Skip
Price report
Sample listing, page not in this demo
The three lines
Reviews
AI tools and agents, installed and used on real work for at least a couple of weeks. Each one gets a verdict, the cost, the licence and what exactly was tested.
ReviewsMarket
The AI token market, once a month: input, cached and output prices side by side, and what a typical month of agent work would cost on each model.
Market reportsHomelab
Self-hosted software on small hardware: what runs well, what needs attention, and the backups I actually restore from.
Homelab notes