Connect
Sort capabilities by what happens when they are wrong.
Reading an analytics figure and changing an access setting are not the same kind of permission, and a flat list of checkboxes treats them as though they were. Four levels let a policy say "nothing above medium without approval" once, rather than naming forty tools.
Low — reading something that cannot be wrong for long
Analytics figures, published content, a campaign's current spend. A mistaken read is a wasted call.
Medium — something a customer will see
Sending an email, creating a draft, booking a meeting. Recoverable, but not invisible: somebody outside the company notices.
High — money, commitments or contracts
Refunds, credit notes, discounts beyond a limit, payment arrangements. These carry mandatory approval, and the approval cannot be configured away.
Critical — access, security and the platform itself
Changing permissions, granting access, altering another agent's configuration. No agent holds a critical capability by default, and granting one requires a second human.
Tools
The rest of this area.
Named capabilities with defined inputs, outputs and risk.
The tool registry
Every capability an agent could be given, in one list.
What a tool declares
Inputs, outputs, risk, permission, approval behaviour.
Your business as a toolbox
Connected systems become capabilities, not open access.
Your own tools
Expose your own service as a permissioned capability.
Adding external systems
The same architecture, more tools. No redesign.
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]