Foldyard
A local, VM-bounded dev environment that holds your whole stack and your coding agent – with no credentials inside the fence
See it in action →
Daniel Demmel
Design technologist and full-stack AI engineer at Tangible, helping green hardware startups scale through debt, not dilution. Climate tech only makes a difference at gigatonne CO₂e scale – I want to help it get there.
When I'm not crunching contracts with LLMs, I'm probably on a side quest with development tooling or down a rabbit hole about metacognition.
More about me→Projects
All projects →
Claude Code Log
Python CLI that turns Claude Code transcripts into readable HTML, Markdown, and a browsable TUI
See it in action →
Online LLM Tokenizer
Browser-based tool for comparing tokenization across different LLM models
See it in action →Recent Writing
All posts →
·18 min read
I thought capturing Linux kernel events would help agent monitoring a lot, but so far it doesn't seem to
I replayed 800 AI agent runs inside a sandbox that records what the Linux kernel sees, and asked a small model to watch them with and without that evidence. A pre-registered test couldn't show it helped – here's what I learnt along the way.
·17 min read
What active parameters don't tell you about running big MoE models on a Mac
··11 min read
Feedback loop engineering
··27 min read