Wispr Flow: Dictation That Works EverywhereStop typing. Start speaking. 4x faster.
Promoted
PureBox.ai
Review-first AI cleanup for your real Gmail inbox.
155 followers
Review-first AI cleanup for your real Gmail inbox.
155 followers
Most inbox tools make you build rules first. PureBox skips that - it's personalized from your first scan using AI and your own Gmail history (which senders you open, reply to, or let pile up), so suggestions fit your routine on day one. Every suggestion shows why, and nothing moves until you approve - no black box, no silent deletions. It surfaces what needs your attention, not just what to trash. Your data is never sold; Pro subscriptions fund it.
Congrats on the launch! The three-bucket plus reasoning approach is a smart way to keep trust intact, most inbox cleanup tools break it the moment they act silently on something that mattered.
Before I'd trust it with my own Gmail, I'm curious what happens to an email's content after it goes through the model to generate the category and explanation, whether it's discarded right after the suggestion is shown or logged somewhere on your end. That distinction would decide whether I connect a real inbox to it.
the "nothing moves until you approve" framing plus the 30 day undo window is the right way to earn trust with something touching a real inbox. one thing I didn't see covered - if someone connects, tries it out, then revokes Gmail access later, is there anything left behind on your end from the scan (cached content, summaries, sender history), or does disconnecting clear it out completely?
Sender history is a strong signal but it breaks in one spot that's bitten me: a reply to a thread I started, from an address I've never opened before, looks statistically identical to a cold email. I run a lot of outbound and the messages I most need to see are exactly those first replies from strangers. Does PureBox weight "this is in a thread I sent" above sender history, or is the classification mostly per-message?
The review-first workflow feels like an important safeguard for inbox automation. Does every cleanup recommendation expose its reasoning and confidence, and can approved actions be rolled back easily?
@shaked_i good to know it stays fully searchable, that was my main worry. follow-up - if someone runs a batch of archives and later realizes the LLM was consistently wrong about one category, like it kept archiving newsletters that were actually still relevant, is there a way to undo that whole batch at once, or is it back to unarchiving one email at a time in Gmail?
Hey I really want to use this! My email box is sooooooooo full! I am afraid I will delete something important.
This is clever. Does the AI learn from rejections too, or just from what you approve?