The notebook

Posts, with verdicts.

Reviews of AI tools, agents and self-hosted software I have actually run, notes from the home lab, and the monthly token market tables. Each one opens with a verdict so you can decide whether the detail is worth your time.

Sample entries. The list shows how posts, tags and verdicts work. Only the gateway review has a full sample page; the other titles and verdicts are placeholders.

10 posts

  1. AI token market, November 2026

    In one line: input, output and cached-token prices per million for the main hosted models, side by side, with what moved since October.

    Monthly12 min read
  2. LiteLLM as a home-lab model gateway: three weeks in front of four providers

    In one line: worth running the moment you call more than one provider; not worth it if you only ever call one.

    Keep14 min read
  3. Two weeks of a terminal coding assistant in a real repository

    In one line: quick and tidy on small, well-described changes; you still have to hold the plan.

    Try11 min read
  4. AI token market, October 2026

    In one line: the same tables, plus a first pass at comparing batch and cached pricing where providers offer them.

    Monthly10 min read
  5. Moving the home lab from Compose files to a self-hosted PaaS

    In one line: nicer deploys, but I gave up more control over networking than I wanted to.

    Wait9 min read
  6. A self-hosted chat front end for the whole household

    In one line: easy to stand up; user management and updates need more care than the README suggests.

    Try10 min read
  7. Leaving a coding agent running unattended overnight

    In one line: it finished the tasks and broke the tests it never ran. Not yet.

    Skip8 min read
  8. AI token market, September 2026

    In one line: the first issue, and the method: list prices only, per million tokens, checked against each provider's own pricing page.

    Monthly9 min read
  9. Self-hosted photo backup, after the first full restore

    In one line: it replaced the cloud album for us; I only trust it because I tested the restore.

    Keep12 min read
  10. Uptime checks and push alerts I actually read

    In one line: fewer, louder alerts; three checks did more than the thirty I started with.

    Keep7 min read