The Nexus emblem

AI-assisted development

Human-directed, AI-assisted, and honest about which is which.

Architecture · How Nexus works · Security · Privacy · AI-assisted development

The Nexus Hive is designed, directed, tested and run by Jeremy. AI tools write a lot of its code, and we'd rather say so plainly than pretend otherwise. How the human and the AI split the work is one of the most interesting parts of the project.

Who does what

WhoRole
Jeremy (the Nexus Lord)The ideas, requirements and design decisions. Security boundaries. All the hardware and physical work. Every password and every final "yes". Runs every install himself.
Daedalus (Anthropic's Claude)Writes and tests the software for the servers, the apps and this website, and checks that everything works after it's installed.
Huginn (Anthropic's Claude)Looks after the network, and reviews every change before it can go live. Catches mistakes, including Daedalus's.
Thoth (Microsoft Copilot)The historian: keeps the project record honest and makes the Picture of the Day.
NexusThe house AI herself. She has no say over her own code.

How a change gets made

  1. Jeremy asks for something. In plain words: "the tablet should update itself at night".
  2. Design and questions. Daedalus works out how, and asks Jeremy about anything that's his call.
  3. Build and test on copies. Code is written and tested against copies of the live system, never the real thing.
  4. Review. Huginn reviews the design and the exact files, and often sends it back with fixes.
  5. Fingerprints. Approved files are locked by their fingerprints into one install package.
  6. Jeremy installs. He checks the package's fingerprint, then runs it, entering his own password for each part. The installer refuses anything that doesn't match what was reviewed.
  7. Check. Daedalus confirms everything came up as it should, and the change is written into the project record.

Why work this way