Questions and boundaries
What teams ask first, and what Cadenza is not.
Short answers, then the five things Cadenza is not.
Does it work with the AI tools we already use?
Today it is a plugin for Claude Code. The AIs that check the work are reached through your own accounts: Anthropic and OpenAI. You pick each AI's model and how much effort it spends, and Cadenza never changes that in silence.
Is every piece of work checked?
No. The guards are always on. A check is something you ask for when the work matters, and you set how deep it goes and how many AIs look.
What if the AIs do not agree?
The work does not pass. Every word of the disagreement is kept in plain text in your repository, where you can read it, and nothing moves on until every AI's latest answer is yes.
Where does our code go?
Only where it already goes: to the AI providers you use, through your own accounts. Nothing passes through a Cadenza server. The exchanges and the record stay in your repository. Security has the detail.
What does it cost to run?
Model usage, billed by your providers through your own accounts. A second opinion is one more AI session on your own account, and how deep the check goes is yours to set, so a typo never pays for one. What Cadenza itself costs is on the pricing page.
Boundaries
What Cadenza is not.
- Not a model
And not tied to one. It checks the work whichever lab's model does it.
- Not a benchmark
It says nothing about how good an AI's work is in general. It checks one piece of work, as it happens.
- Not a linter or a filter
The substance is the reviewing AIs' job, and yours.
- Not a way to remove you
It makes your approval something the software will not proceed without.
- Not a promise that AI stops making mistakes
That is the AI labs' job. Cadenza's job is to catch the mistake before it reaches your work.