Finance Chief

Once a day, Warden runs it to total the month’s AI spend across the apps from Warden’s own usage log, and it reports what it found in the audit log. Read-only: it watches spend, it never spends.

Sorts new transactions into your spending buckets using the rules you’ve already set — on its own, no approval needed. When a transaction matches no rule and it can’t confidently place it, it stops and asks you instead of guessing.

Turns raw numbers into plain-language observations about spending. It summarizes patterns via Claude and suggests — it never changes your data.

Answers your questions about your own finances, grounded in your private ledger. It runs on its own, but only ever reads — it can’t move a number.

Takes a second look at large “spent” transactions and flags the ones likely filed in the wrong bucket — a pass-through that’s really internal, an asset buy that’s really invested — with a proposed bucket and its reason. It proposes only: every reclassification waits for your one-tap confirm.

Values the India flat by searching the live market for current ₹/sqft and multiplying by its area, then writes that into net worth. It runs on refresh with no prompt — the most autonomous agent here, and the only one that reaches the open web — so it’s rate-capped and now fully metered.

Drafts a fourth path from a conversation — forks a scenario and changes only the fields you named, with a required rationale. It never applies anything: a drafted scenario is a preview that waits for your explicit approval before it’s saved.

When a role in another city reaches the interview or offer stage in Career Chief, it tells Finance Chief: company, role, city, stage and the salary as written, nothing else. Finance Chief then asks her whether to explore that city as a path. It never sends notes, mail or briefs, can’t read anything in Finance Chief, and doesn’t notify her itself.