iPhone app · prototype, not yet released

Most recordings look backward.Yours will move things forward.

A meeting. A half-formed idea in the car. A passing “we should do that sometime.” Record it and walk away.

Where other tools stop, and where this one goes Other tools go from recording to summary and stop. This one continues to the thing happening or not happening, and returns to update the original note. EVERYONE ELSE Recording Summary stops here THIS ONE Recording Summary It gets done Or it doesn't either way, the note you recorded gets updated
Otter, Granola and Fireflies all stop at the summary. Everything after that was always your job — including remembering, three weeks later, whether it happened.

I've been running this on my own life for months. I'm now rebuilding it for someone other than me.

Join the advisory round Four questions. We're deciding what to build first.
↓ Skip to the eight things it does
01 — one recording

One thing said, four places it belongs

Four sentences from a twenty-minute meeting. They aren't the same kind of thing, so they don't go to the same place.

The third row is the whole argument. Every other tool turns that sentence into an action item. It wasn't a decision, so it doesn't become one — and you can still open utterance-17 to see exactly what was said.

It doesn't have to be a meeting. A stray thought in the car takes the same path, and comes back later when something connects to it.

You could paste a transcript into ChatGPT and get a decent list. That's a tool you have to open. This one writes to your calendar unasked, and comes back on Friday to say whether the thing actually happened.

03 — eight things · tap to open

What it does with all of it

Which of these you'd actually use is the thing I'm trying to find out.

Flags conflicts instead of guessing

Overlaps and contradictions get raised, not quietly resolved.
Most assistants resolve ambiguity silently and hand you a confident answer. It's sometimes wrong, and you never find out which times.
calendar collision“Tuesday 3pm overlaps something already scheduled. Nobody mentioned it. Which one moves?”
contradicts an earlier decision“This reverses what was decided on July 22nd. Is today's version current?”

Done comes back

Check it off, and the note it came from updates too.
Tasks land in the tools you already use. When one is finished or dropped, that travels back to the original transcript — so a three-week-old note tells you what happened, not just what was said. This is the loop everything else is built on.

What other people promised

Not your to-do list. Your to-check list.
Every meeting tool extracts your action items. What others committed to lives only in your memory, which is why you don't follow up on Friday. Tracked separately, by owner.

Briefed before you walk in

The context you'd have looked up, already pulled together.
What was decided last time, what's still unanswered, what you promised and haven't done, and which of those they're likely to raise. You don't ask for it.

A page for every person

What you owe them, what they owe you, what's unanswered.
Built from every conversation they've appeared in, not a contact card you had to maintain. Months of context on one screen, each line linked to the moment it came from.

Where a project actually stands

What's settled, what's open, what nobody answered.
One page per project that doesn't go stale between meetings, with an honest marker showing which conversation it's current as of.

Old thoughts, resurfaced by relevance

Not on a timer — when something you're doing now connects.
Three weeks ago you said onboarding might work better as an interview. Today a meeting turns to new-user drop-off. It links the two. Search can't do this, because search requires knowing there's something to look for.

Find it the way you remember it

“That thing I said around 3pm yesterday.”
People remember by rough time and fuzzy topic, which is what folders are worst at. Ask in plain language and results point at the exact moment, not a keyword match.
04 — why trust it

Built by someone this has already burned

A real incident from running an early version on my own life: a priority the assistant had inferred got stored as though I had decided it, then quoted back to me for weeks. A position I never took became something the system took for granted.

So observation, inference, proposal and decision are four different things in the record now, and only an explicit action of yours becomes a decision. Repetition doesn't promote anything.

Records are plain Markdown and export in full. Disconnect anything, delete recordings, delete the account. Nothing sold to advertisers or used to train models by default.

05 — honestly

Where this actually is

running daily · one user, meCapture, transcription, meeting notes, task creation, completion flowing back, decision history, open questions, project status.
being moved into the app nowThe iPhone app, the approval screen, search.
nextContradiction detection across meetings, and the person pages.

It'll be paid eventually — around a couple of coffees a month. Nothing to pay now, no launch date to announce.

Join the advisory round No spam. You'll hear from me directly, or not at all.