Your life story, one chapter at a time

The new Life Book tab is where it happens. Remember This looks at what your library already knows about you — where you've lived and traveled, who shows up in your photos across the years, what your voice memos talk about — and splits your life into eras. No setup, no questionnaire: the eras are detected automatically from your own archive.

Pick an era (the most recent one is pre-selected) and hit "Write a chapter." Claude drafts one focused chapter for that span, in the background, in a few minutes. While it writes, it asks you the two to four questions only you can answer: who's in the 2019 photos, what actually mattered that year, whether a trip was work or a vacation. Answer right from the question card, and the chapter folds your answer in before it's done.

Pick the same era again anytime to add more detail — each pass builds on the last. And if you'd rather see the whole arc at once, "Draft whole biography" runs a single holistic pass over everything.

Knowing what's happening, and correcting it

An AI writing about your life in the background could easily be a black box. Life Book isn't: it shows an honest status the whole time — is Claude still writing (step away), is it waiting on you, or is it done. Open questions for the current chapter sit right there in the tab, and a free-form note field lets you tell the running session something it didn't think to ask — "that trip was actually 2018," "skip the work stuff."

Corrections you give are remembered durably and folded into every future chapter, not just the one you're looking at — so Life Book gets more accurate the more you use it, instead of repeating the same mistake in every era.

Answer by voice

The question inbox — in Life Book, and anywhere Claude asks you something mid-background- session — now has a push-to-talk button. Speak your answer instead of typing it, transcribed locally and instantly, with a live level meter so you know it's hearing you. A recorded answer is never lost: if transcription hiccups, the clip is kept and Retry resubmits it rather than silently discarding what you said.

The microphone works again

A missing hardened-runtime entitlement was silently blocking microphone access on every brand — no permission prompt, no entry in Privacy & Security, nothing to toggle. It's fixed. Push-to-talk answers and in-app recording now request and get mic access properly, and a firm trackpad press on the record button no longer triggers macOS Look Up mid-recording.

Smaller things

• Transcription reliability fixes. New recordings start transcribing the moment they're detected instead of waiting for the next scan, audio is enhanced first for better results on quiet or noisy recordings, a transcript that goes off the rails (e.g. a hallucinated repetition loop, now detected) is flagged with a quality note instead of being trimmed, and a long transcription no longer restarts from scratch after an app restart.
• Re-transcribe in the right language. A voice memo that came out in the wrong language can now be re-transcribed with the language forced, from the row's context menu.
• New photos are caught within a day, guaranteed. The photo watcher now runs a daily full-scan safety net on top of live change detection, so a photo that slipped past the live watcher (iCloud sync timing, Mac asleep) still shows up.
• A resource history chart. The Processing tab now graphs the daemon's memory and CPU over time — "is it leaking, what's it doing right now" is answerable at a glance.
• Background sessions report when they're actually done. A finished chapter or dreamer session used to sometimes show "Running…" forever; status now distinguishes "still writing" from "waiting for your answers," and "Join session" on a question card reliably joins the session that asked it.
• Lighter on memory and CPU. Fixed a daemon memory leak in photo processing, bounded the captioning engine's footprint, cut the standing CPU of the library reconciler, and made thumbnail encoding markedly faster.
• The CLI imports into your library. rememberthis transcribe now routes the file through the daemon by default — copying it into your Remember This folder, creating the registry note and transcript, and streaming live progress the app picks up — instead of just printing text. Pass --text-only for the old behavior.

What's next

This is the first release of the life-story feature, so the next round is mostly about quality — watching how chapters read across different eras and tightening the writing where it doesn't hold up. If a chapter gets something wrong, the note field is the fastest way to tell us, and it'll be right the next time.