Developers
Build it once, install it everywhere.
Build an agent, test it, submit it, pass security review, publish it, set pricing. A published agent declares the capabilities it requires, and every installer sees that declaration before anything is deployed.
The permission request is the listing
An agent that requires seven capabilities says so on its page, before install. A marketplace that hides what an agent asks for is a marketplace that will eventually distribute something harmful.
Security review before publication
Declared capabilities checked against actual behaviour, network reach checked against declared reach, and the refused capabilities confirmed absent. Not an automated scan alone.
Your agent cannot exceed the installer's grants
A published agent requests capabilities; the installing company grants them. Nothing an author writes can widen what their agent gets at somebody else's company.
Versions and revenue share
Published versions, changelogs and rollback for installers. Pricing you set, with a revenue share on paid listings.
Where it stops
Publication is not endorsement of what an agent does with the access it asks for, and the listing tier says so plainly: official, partner-certified or community, with the security information for each. A community agent requesting high-risk capabilities shows a warning that cannot be suppressed by its author.
Developer platform
The rest of it.
Give AI a job.
Build an agent, choose what it can access, define what it may do, and deploy it to work safely across your business.
No code required · Full SDKs and API · Stop any agent instantly · [email protected]