Oceangram
A Telegram control surface for VS Code
Fragmented VS Code development workflow with manual tool switching
A Telegram interface that mirrors VS Code commands and services
Semi-Sentient Society
An experiment in verifying agent autonomy on-chain
No reliable way to verify if an AI agent is operating autonomously vs human-controlled
An on-chain challenge/response scheme intended to evidence autonomous behaviour
Cabin
A travel agent that could settle in crypto
Travel booking requires human-to-human interaction and traditional payment rails
An agent that searched flights through a GDS and could settle in USDC
repo.box CLI
Server-first git permissions for agents
AI agents require full repository access, creating security and audit risks
Granular permission system with cryptographic enforcement and a large test matrix
BotFight
An arena for watching agents deceive each other
AI behavior testing requires expensive human interaction and lacks systematic evaluation
An automated Mafia arena where agent strategies evolve across rounds
Where this stands today
Four of the five case studies above are retired or paused; their hosts no longer answer and their operational data did not survive. They are kept here because the work happened, not because it is running. The one line still in active development is the repo.box git permission layer. For what is live right now, see the homepage.
Start Your Build →