Verdict is an AI development platform that investigates your repository, fixes the problem, and proves it works — tests passing, cost visible, done. Routine work runs free on your own machine. Frontier cloud models are one click away when the problem deserves them.
Windows coming soon.
Give Verdict a failing test, a stack trace, or a bug report. Watch it work the problem end to end — and check every step for yourself.
Start from a red test, a stack trace, or a bug report. Verdict reproduces the failure before it changes a line.
It reads the repository and follows the call path, showing every command it runs to understand what actually broke.
You get a plain-English explanation of why the bug happens — not just a patch you're asked to trust.
A tight change scoped to the problem — not a rewrite of every file it happened to open.
A new test that locks in the fix, so the same bug can't quietly come back later.
The suite runs again and comes back green. What reaches you is finished work with the proof attached.
You get finished work, not homework.
verdict-code runs with zero account and zero config, and MVAI ships a capable local model out of the box. Turn off Wi-Fi and keep coding.
Install verdict-code and go — no sign-up, no API key, no cloud round-trip for routine work.
Run models on your own machine, a box on your network, or your own provider key. Your choice, every request.
Escalate to frontier cloud models for the hard problems, then drop back to free local work. One product, both worlds.
Verdict starts at $0 and stays there for local, LAN, and BYOK work — no seat fee, no metered floor to escape.
474 tok · 53 in / 421 out · 51.3 tok/s · 8.3 s · 0.0829 Wh · Free (local)One real response from our 4B model on an Apple-silicon laptop.
Every successful request shows its real cost — tokens, speed, time, and for local models, energy. Cloud credits settle at real vendor cost. No mystery meter, no surprise bill.
One credit balance, projects synced across everything you use.
A full macOS app with chat, projects, and themes, powered by the bundled local stack.
The standalone surface — works with zero local install, cloud-direct, straight from your terminal.
Bring Verdict into the editor you already use, with the same models and the same billing.
Reach your own machine from the browser over a secure tunnel — your code stays home.
90+ certified cloud models — plus every local, LAN, and BYOK model you can point us at. Our own Verdict models are in the catalog for chat, too.
An in-product benchmark lab with 25 suites built in lets you test any model on your own hardware before you rely on it. Measure it yourself — speed, cost, and quality on your tasks.
BYOK runs at cost: bring your own key and pay your provider directly.
We never train on your code. There is no toggle, because there is nothing to turn off.
Run models on your own machine and your code never leaves it. Privacy is the default, not the upgrade.
Remote access reaches your own machine over a secure tunnel. We don't keep a copy of your repository.
Start free — local, LAN, and BYOK models cost nothing, forever. Buy credits only when you want cloud.
Code all day for $0 — local + LAN + BYOK, no cloud required.
Verdict for solo developers
Verdict for power users
Collaborative AI development for organizations
Start free with local models. Add cloud credits only when a problem is worth it.