Skip to content
sgit/ store
Produced with model assistance. Every page on this site was drafted with a large language model and reviewed by a person before publication. Nothing here is a compliance assessment. What we do not say, and why How every claim here is evidenced v0.3.18

Describe your agent

Make its reach visible.

Choose what your agent can do, out of the 23 things an agent can be granted. What you build stays in this browser and leaves as text you copy out or a file you download; nothing here is sent anywhere.

What a grant is, and what it is not

filesystemfiles and directories

processprograms and their execution

networkendpoints and hosts

identitycredentials and who the agent can act as

communicationmessages to people

coderepositories and what lands in them

moneybudgets and spend

schedulethings that outlive the turn

browserwhat a browser extension or automation can see and do in your browser

What this produces, and what it does not

It produces the grant — everything the agent can do, as you describe it. That is one of four objects in an Agent Behaviour Policy. It is not the mandate, not the delta between them, and not a policy. It is the half you can answer from the deployment, which is why it is the half given away; and it is described by hand, so MAP-A-GRANT.md, which ships in every template pack and measures the same thing from the deployment, will disagree with it wherever somebody guessed.

THE HALF THAT NEEDS A PERSON

ABP Tailored, £500

A named security professional corrects the mandate against your details. The document this page produces is one of the two files you send back; the corrected vault follows 1 to 3 days from your reply.

delivered by a person ↗

Buy this level

THE VOCABULARY IS NOT OURS

23 primitives on one grammar

verb.object.reach, promoted from riskmandate.ai’s own template vault so that what you describe here is in the same words as the document you would receive. Nothing in it was written by this store. The fifteen published shapes