AI agent platform administration: the parts you only notice when they are missing
Who can change what, whose instructions every agent follows, what language the person in front of it reads, and who gets told when something happens. The operational layer that makes this a platform you can run day to day.
Roles you configure, identity that is managed
A business is not one person. Roles (administrator, editor, viewer and support) come with a permission matrix you configure and not one you accept, and identity is managed: sign in with your existing work account, with a second factor where you require one. Every external link the platform issues is short-lived and signed.
Standing instructions for every agent
Above all of it sit your standing instructions: a versioned library of what every agent is told, seeded by a questionnaire when you start, previewable before it applies and rolled back in one move when an edit turns out badly. Support staff can act in your workspace only through guardrailed, audited impersonation, which is visible to you.
What sits underneath
The operational layer
Where your data lives, who can reach it and what is on the record is its own page. This one is about running the platform day to day.
Roles and identity
Administrator, editor, viewer and support, with a permission matrix you configure. Sign-in through your existing work account, with a second factor where you require one.
Your standing instructions
What every agent is told, in one versioned place, previewed before it applies and rolled back in one move when an edit misfires.
Notifications
Email, an in-app feed and mobile push, with preferences, so the person who needs to know finds out and the rest are not buried.
Five interface languages
The dashboard, portal, widget and extension ship in English, Spanish, French, German and Portuguese, and an agent replies in the language the person wrote in regardless.
Billing and allowances
Plans, add-ons, metering per dimension and threshold alerts, with a billing portal, plus commissions, statements and payouts where you refer work to us.
Health you can watch
Tracing, metrics and logs per organisation and per run, with readiness checks, so a problem is something you can see and not something a client reports.
Change how every agent behaves, in one place, reversibly
The instruction that matters most, what your business never promises, should not have to be typed into each agent and remembered on each new one. It is written once, applied to all of them, versioned, and reversible in a single move when it turns out to have been the wrong wording.
- Seeded by a questionnaire when you start, so the first version is not a blank page.
- Previewable before it applies to anything a client can see.
- Rolled back in one move, which is the part that matters at the moment you need it.
What you use it for
Start with the rule you never want an agent to break
Client service & accounts
A named client asking where their own account stands, and waiting for a person to look.
Read moreBack-office automation
Renewals, closings and filings that run on a calendar, not on somebody's memory.
Read moreWhere your data lives is its own page
This one is about running the platform. For EU residency, your own encryption key and what is on the record, read the Trust page, or bring the questions to a demo.