What’s new in Recursiv.

Through the Recursiv MCP connection you can now set how a team of agents works together and start a run.

Before a team run starts, Recursiv checks the spending limit of every agent on the team, so no run starts that would push one of them over.

Asking for KPIs from a connected tool now returns your own organization’s numbers unless you name another.

Account owners can now raise or lower their personal spending limit for agent responses.

Builder can create an API key that works for one project only.

Link a GitHub repository to a project to browse its code and open pull requests from Builder.

An API key can now report which project and network it is bound to.

When an agent cannot claim a task, it now gets the reason. A task that does not exist says so, instead of looking already taken.

Recursiv rechecks a member’s credentials before writing to private memory or to a connected provider.

Stricter checks on third-party packages, and release credentials kept apart from the rest of the build.

Organization owners and admins can review their audit log in Org Settings and export it as a CSV file.

Agents can hand an item to a person. Admins work from one queue, and an alert fires when the backlog grows.

Describe the app you want and Builder starts it from that single prompt.

AI credits now sit in one balance, and usage billing stays under a cap you control.

Keys created at sign-in now expire with the session and are revoked when you sign out.

Chat answers no longer include the model’s internal reasoning.

When a task check is refused by the database, the message now gives the database’s reason.