One neuron at a time
by Caleb
Push single neurons in small open models and turn them into villains, flatterers or pirates. Try the recorded replies, then run it yourself.
A guided tour of Dave, a personal AI agent: what it can do, why each feature exists and how it is put together.

Dave is a personal AI agent that works for one person. It runs all the time on its own small server, built on Anthropic's Claude Agent SDK, so it has the tools a coding agent has: a shell, files, web search and plug-in tools. Its owner talks to it from a phone or Mac through chat apps.
This site documents Dave feature by feature. Each page says what a feature does, why it is there and whether it is off by default. A big theme is safety: how to let an agent do almost anything without handing it risky power.
A Build your own section sets out the architecture and build order for people, and coding agents, who want to make a similar agent. Every page also has a Markdown copy for bots.
by Caleb
Push single neurons in small open models and turn them into villains, flatterers or pirates. Try the recorded replies, then run it yourself.
by Jeremy
Capability-based security, scope-bounded planning and out-of-band audit for autonomous LLM agents.
by Chan
An open-source language for floor plans, now also inside ChatGPT: ask for a plan in plain words and ChatGPT writes it in ArchLang, compiles it and checks it.