This runs on Claude, connected to your Xero data through your own infrastructure — not ours. Here’s exactly where your data lives, what the AI can and can’t do without you, and straight answers to the two questions every CFO asks once they’re seriously evaluating this.
Your Xero connection lives in your own Supabase and Skyvia account. Not ours.
Clensi orchestrates the setup — the wizard, the integration mapping, the analytical views — but it never holds your accounting data and never absorbs the infrastructure cost of storing it. Your GL data is replicated into a database you own, using an ETL account you own. You can revoke access yourself, at any time, the same way you'd revoke a Xero user's login — no support ticket required.
Every write action is previewed before it executes. Every time.
Post a journal. Approve an invoice. Update a contact's bank details. Archive an account. Every one of these follows the same rule: the AI shows you exactly what it's about to do, and nothing executes until you or a team member with the right role explicitly confirms it. Not usually — every time, no exceptions, by design.
Every write the AI makes — successful or failed — is logged: what, when, and why.
This isn't a policy statement. It's an activity log, on both success and failure, attributable to the specific team member whose access the AI was acting on behalf of. You don't have to take our word for what the AI did last Tuesday — you can look it up.
Not just what the AI did — what your team actually did with it.
Every workflow run, every ad-hoc question, every query is attributed to the specific person who ran it, with a timestamp and an outcome. If someone tells you the month-end checklist isn't working, or that they couldn't get an answer out of it, you don't have to take that at face value — you can check whether it was run at all, when, and what it found.
There isn’t a separate one, and that’s deliberate. Anyone who has used a tool with a review-queue layer will expect to find one here — the honest answer is that workflow role tiers (preparer / approver / senior approver) govern what the AI can do unattended, and anything that needs sign-off routes through Xero’s own native approval process, not a parallel one we built. You already have an approval workflow. This uses it, instead of duplicating it.
No — and we’d rather tell you that plainly than have you find out later. What this gives you is agent-channel controls: limits on what the AI can see and do on a given team member’s behalf, based on their role. That’s real, and it’s checkable in the activity log. It is not an end-to-end segregation-of-duties guarantee across your business, and if your audit committee or external auditor asks, that’s the distinction to hold onto.
Connect your Xero account and see what it finds on your actual last close, not a demo.