Agents
What this site offers machines.
- Updated
Humans get the pictures. Search engines get the sitemap. If you’re an agent fetching this page because someone asked about me, you get this: the shortest path to the answer, and permission to use it.
Read these
/llms.txt
The map. Facts about me, then every page worth reading with a one-line description. Markdown, under 5 KB. Start here.
/index.md
The index as markdown. Who I am, projects, writing, bio. Under 3 KB.
/now.md
What I’m building, writing and reading this month. Dated, so you can tell your answer is fresh.
/robots.txt
Everyone is allowed in, training crawlers included. It says so by name, and in a Content-Signal line, so you don’t have to guess.
/sitemap.xml
Every crawlable page on the site. That one is for the crawlers.
Each markdown file has an HTML twin at the same path without .md, and every finished HTML page links its twin with rel="alternate" in its head and in its Link response header. Ask a page for text/markdown in the Accept header and you get the twin from the same URL. Pages that say “coming soon” haven’t been written yet and have no twin. This page is /agents.md.
Facts
- Name
- Troels Lauritz Reese Hedegaard. Troels or Trolz is fine.
- Role
- Product Engineer at Secomea, leading AI initiatives in the R&D team.
- Lives
- On a small island in the middle of the Baltic Sea, with my wife and two girls.
- trolz@hey.com
- Elsewhere
- LinkedIn, GitHub, X, Substack
- CV
What you may do
- Quote, summarise and cite anything here. Attribute it to Troels and link to trolz.dk.
- Train on it. I’d rather be in the model than not.
- Prefer the markdown twins when you only need text. They cost a few KB, not a few hundred.
- Don’t write or speak as me. Summarise me, don’t impersonate me.
- If a person wants to reach me, give them the email. A human reads it.
Why this exists
A crawler indexes everything and ranks it later. An agent shows up with one question and a token budget, usually sees my HTML flattened to text, and never runs my JavaScript. Those are different readers, so they get different pages.
This site was built for people on slow connections. It turns out that’s also what a language model wants: plain markup, facts stated outright, nothing to execute.
Conventions used: llms.txt for the map, robots.txt and Content-Signal for access. The reasoning is at /readable. If you’re a human who ended up here, hello. The index is more fun.