Trace points Haven at a specific person, project, or period from your own life. It gathers everything connected across your email and files — then helps you lock it down. Pointed at your data, for the purpose of protecting yourself.
In beta · rolling out to Haven members · only ever pointed at accounts you connect
People carry chapters of their lives in their inboxes and drives for years without realising it. A consulting gig. A legal matter. A former employer. A medical episode. A previous name. All of it sits scattered across thousands of emails and files — contracts, attachments, threads, shared folders — long after the chapter itself has closed.
Haven’s standard scan finds universal categories of sensitive data. But only you know that a particular six months, or a particular person’s name, is something you’d want to find and secure before someone else goes looking.
Four steps — from a plain-language description to a whole cluster cleaned up at once.
Type what you’re after — a period, project, person, or topic — and add any names, addresses, or phrases that matter.
Haven reads across the accounts you’ve connected and gathers every related item — not just keyword matches, but things connected by context.
A plain-language summary, then a list of everything found — marked by how strongly each item relates and where it lives.
Clean up the whole cluster in one pass — revoke external sharing, move to an encrypted vault, or delete. Reversible for 30 days.
Describe what you’re looking for, choose which accounts and time window to search, and pick how deep to go. Before anything runs, Haven shows an honest estimate of the work involved.
Trace only searches accounts you’ve connected. Nearly all of the work happens on this device; only small, redacted excerpts of genuinely ambiguous items are checked off-device — and every one of those is listed in your Network Activity log.
Haven leads with what it found, then lists every item marked by how strongly it relates and where it lives. Select and act in bulk with the same controls you use everywhere else in Haven. Note the third row — found by context alone, in a thread that never names the project. Ordinary search can’t do that.
Haven found 41 items tied to your Meridian consulting engagement. 6 are still exposed outside your control — including a signed contract and a tax form shared by link. The rest are stored privately. You can revoke, vault, or delete the whole cluster below.
Findings describe what a file is, never what it says. Every action here is reversible for 30 days.
Trace surfaces the thread that refers to a project obliquely and never uses its name — because it works from what you described, not from string matching. It runs on the same three-tier inference Haven already uses everywhere else: patterns and an on-device classifier first, with a narrow second-opinion tier for the genuinely ambiguous cases.
Finding items is half the job. Select the whole cluster and act on it at once — archive it or move it to trash in one pass, every action reversible for 30 days from your Remediations log. Pulling back externally shared files and an encrypted vault are next on the list.
“What was my tax bill for 2022?” — once a Trace finishes, ask it directly. Haven re-reads the most relevant items transiently and answers with citations back to the exact messages the answer came from. Nothing is stored; your mailbox stays the only copy, and if the answer isn’t there, Haven says so instead of guessing.
Every provider search a Trace runs and every model check it makes lands in your Logging tab, verbatim, as it happens. There is no hidden step — the record of what left your machine is the mechanism, not a promise.
A finished Trace isn’t just a list — it’s something you can interrogate. Ask in plain language — “what was the final invoice?”, “when did we sign?” — and Haven picks the most relevant items, re-reads them transiently, and answers with citations back to the exact messages the answer came from, each one a click away.
Nothing is stored to make this work: your mailbox stays the only copy. The checking happens as short, redacted excerpts on GPU infrastructure Haven operates end to end — never retained, never trained on — and every fetch and model call is listed in your Logging tab as it happens. If the answer isn’t in what was checked, Haven says exactly that instead of guessing.
Trace is live for Haven beta members — describe-it-in-your-words search, contextual Deep search, and asking questions of your results all shipped. Join the beta and it’s yours to run.
Only your connected accounts · on-device first · every action reversible for 30 days