Contact
Get in touch.
LinkedIn
GitHub
Email
A set of tools to give your coding agents context as a live architecture diagram.
Your agent reads files and guesses. Give it a live map of the whole system instead — what the code is, and what it does when it runs. Context it can trust.
see the model
Anything else: instrument it with /neat extend, or point your own OpenTelemetry exporter at NEAT.
Write your team's rules down once, and every agent stays inside the lines as the system grows. Dynamic guardrails as plain JSON rules — precise, concise, and cost-effective.
# policy.json design-system @neat/ui ports api:8080 · mobile:8081 frontend next.js only backend fastapi → sandbox(java) → iterate data only billing, orders → postgres
Your code declares one thing; production does another. NEAT shows both — a call it makes silently, or one it claims but never runs — and an agent works the gap.
NEAT remembers every state of your code and your traffic. Walk them back, and an agent finds exactly when a bug began — and why.
When something breaks, walk the live edges to everything downstream that's hit — ranked by how sure NEAT is.
Soon: an agent proposes a feature as a sandboxed experiment, checks its blast radius, and ships to main only when it's safe. Agents that build like product engineers.
Blind, an agent brute-forces: grep, guess, edit, retry — burning tokens to make up for not seeing the system. A bigger model is the same move, scaled up.
NEAT gives it ground truth instead. The answer is a traversal, not a search. A lookup, not a hunt.
So you stop paying tokens to compensate for blindness. Not just more accurate — cheaper.
NEAT is free and open source. Run npx neat.is on your machine or your own box and you're live in one command. Want hosted, multi-tenant NEAT? Leave your email and you'll be first.