Posts
Reviews come from running the tool on real work for at least a few weeks. The market post goes out at the start of each month. Homelab notes cover what runs on the servers under my desk.
Showing 10 posts.
-
AI token market, November 2026
Input and output prices for the main hosted models side by side, with what changed since October.
Monthly table -
Open WebUI, six weeks as a small team's chat front end
Verdict: worth running if you already host models; read the licence's branding clause first.
Run it -
Aider in a real repository
Verdict: good at small, well-scoped diffs. I still read every commit it makes.
Run it -
Moving the homelab from Compose files to Coolify
Verdict: a pleasant dashboard, but I missed reading the whole stack in one file.
Try it -
AI token market, October 2026
Same table, one month earlier, plus a note on how cached-input pricing changes the sums for agent workloads.
Monthly table -
AnythingLLM for document questions on a laptop
Verdict: quick to start; answer quality depended far more on how I split the documents than on the app.
Try it -
Letting a coding agent fix CI overnight
Verdict: it fixed the easy failures and faked a fixture for a hard one. Not ready to run unattended.
Skip -
AI token market, September 2026
The first table in this series, and how I decide which models make the cut.
Monthly table -
Immich replaced my photo cloud. Backups are the real work
Verdict: the app is the easy part; plan your restore before you delete anything else.
Run it -
Ollama on a mini PC: what fits in 32 GB
Notes on which model sizes stayed usable on a small box with no graphics card, and where it stopped being worth it.