Low, balanced, max
Every workflow ships with three profiles. What each one changes, how the price is computed, and when each is the right choice.
// about
Open by default_
Most AI security scanners are black boxes: you get findings, not the prompt, the tools, the model or the benchmark. Midkernel is the opposite. The workflows are open source and versioned. Their scores on public benchmarks are published continuously. You connect a repository, pick a workflow whose source and bench row you can read, choose how deep it should look, run it in an isolated sandbox, and read the whole log. You pay for the run, in credits, at a price you saw before you started.
// the story
Midkernel started as a tool for our own work: running open-source security workflows against bug-bounty targets without rebuilding the plumbing every time. The plumbing — sandboxes, model access, artifacts, and a way to tell which workflow was worth running — turned out to be the product.
The workflows were never the secret. The best ones already lived in public repositories, written by researchers and auditors who wanted them used. What was missing was a place that mirrored them, pinned them, scored them on the same benchmarks, and ran them for you at a price you could see first.
So that is what we build: a registry, a bench, and a button. Everything that can be open source is. The company earns the difference between what a run costs and what it is worth to have it be one button.
// founded2026
// based inNew York, NY
// fundingself-funded
// teama small research team
// what we believe
Show the log.
A finding without its trace is an opinion.
Public over private.
Workflows, benchmarks, the runner, the methodology. Closed only where a customer's code requires it.
Price first.
You see what a run costs before it starts, and the invoice matches.
Small on purpose.
A research team, not a sales floor. Growth comes from being the obvious way to run open work.
// research
Pay for runs, not seats
Why Midkernel is priced in credits, why packs never expire, why there is no free plan, and what happens to your money when a run fails.
The harness matters more than the model
Two teams point the same model at the same code and get different findings at a different cost. The difference is text, and text can be open.
// press