Learnings / Jev
Jev: the new decision model on the block
We used Jev to assess AI agent skills. Explore the WhichSkills experiment, its findings and the evidence behind the judgments.
Read the article48Nauts / Software, research & learning
Made in SwitzerlandUnderstand what your AI does.
Decide where your data goes.
Keep evidence of what happened.
I build software that makes this possible.
Starting with NautGate.
48Nauts / Software, research & learning
48Nauts investigates how AI systems behave, what they expose, what they retain and what they cost. We turn that work into practical software, research and guides for building AI-driven applications and services with clarity and control.
xNaut / Agentic development
Keep agent work, worktrees, changes and human review connected in one workspace.
Bowden-PII / Data minimisation
Replace supported personal identifiers locally before an application sends text onward.
Engram / Organisational knowledge
Connect sources, context and earlier decisions before asking a system to propose the next step.
NautGate / AI oversight
Follow recorded model use, routes, tokens and spend for the applications and agents you connect.
More products & experiments
Inspect connected AI requests, control model routes and keep verifiable records.
Connect sources, context and earlier decisions to prepare evidence-backed proposals.
Tickets, parallel agents, isolated worktrees and verified handbacks, with a human final review.
Replace supported personal identifiers locally before your application sends text onward.

Learnings / Jev
We used Jev to assess AI agent skills. Explore the WhichSkills experiment, its findings and the evidence behind the judgments.
Read the article