New: AI token market, November 2026
Full-stack and DevOps contractor
I’m Lance Keay. I build web apps and the servers under them, and review AI tools and self-hosted software after running them on my homelab.
The four latest posts
Sample posts that show the format. Each one ends on a single verdict, with the cost and the licence beside it.
- AI token market, November 2026Market
Price per million tokens in and out, model by model, in one wide table.
- Ollama and four other local model runnersReview
Start with Ollama. Move to vLLM once several people send requests at once.
- Aider in a real repositoryReview
Good at small, well-described changes. Read every diff before it commits.
- Paperless-ngx for a household’s lettersHomelab
Worth the evening of setup. Back up the media folder, not only the database.
How each review gets tested
-
1. Install from the public docs
No vendor demo account. I follow the README on a spare VM in the homelab and write down every step it left out.
-
2. Run it on a real job
It goes into a job I would do anyway: a client repository, my own notes, this site’s build. Toy prompts hide most of the problems.
-
3. Publish the cost, licence and verdict
Every post opens with what it costs, the licence, the version tested and one line on whether to use it.
Contract work. Apps and infrastructure.
- Web apps, end to end
From the database to the screen, with tests your team can keep running.
- Pipelines and infrastructure
CI/CD, containers and infrastructure as code, written down so your team can run it.
- Self-hosting
Moving tools from paid SaaS onto servers you control, with backups that have been restored once.
- AI tools in your workflow
Choosing models and agents by cost and by how often they fail, then wiring them into real work.
