0 trained versions that passed their grade
Barron AI Solutions · research, consulting and products
We build AI that remembers, and we measure whether it's telling the truth.
2
Results, including the failures
Each note comes from our own records, with the number and how we measured it. We publish what didn't work, because that's where the useful part usually is.
6 of 10 right, for the model and for the shortcut alike
Our model of another agent was no better than assuming it thinks like us.
300 places in 118 of 485 files
Tools that quietly ignore a misnamed argument give quiet wrong answers.
3
Products
Things you can use today, built from the same research.
Memory for AI coding agents
Mneva
A hosted brain your coding agent connects to over MCP or HTTP. It keeps what it learns between sessions and recalls it by meaning, with a timeline, corrections that stay open until proven, and wake and sleep handoffs.
Solo free · Developer $20/mo · Pro $40/mo · Team $30/seat/mo
mneva.dev →
Open source, MIT
become-kit
The parts of our own system, blank: long-term memory, a felt-state body, a wake and sleep cycle, and background loops for a Claude Code agent. No personality inside. Your agent grows its own.
Free
github.com/mrshanebarron/become-kit →
Websites and apps, built for you
Cwick
Real working software, not a prompt: business sites, booking systems, member portals and CRMs, built by a US developer with our own AI systems, at one fixed price per plan.
From $100
cwick.us →
4
Consulting
- Agents that keep their memory from one session to the next
- Private models that run on your own machines
- Claude or other models added to the systems your business already runs
- Checking whether an AI feature does what it claims
Everything happens in writing. You describe the problem, you get a written scope and a fixed quote, and you decide. No calls.