Record now, transcribed next

The headline change in v0.3 is the one you'll feel immediately. Hit record on your phone, your watch, or your Mac, and the new memo lands at the front of the transcription queue — not behind whatever your iCloud Voice Memos backlog happened to be when you opened My Transcriber for the first time.

Two bugs combined to bury new recordings under older ones. The scheduler tiebroke FIFO inside each priority bucket, so a fresh memo joined the end of a long line. And the scanner had a "skip enqueueing if the queue already has N items" guard, so new arrivals could fall off the enqueue path entirely until things drained below threshold. v0.3 tiebreaks by capture time and drops the cap-skip. New memos surface within seconds and start transcribing next.

Pending memos in the Transcribe and Timeline tabs now show a "Queued · #N" position pill so you can answer "is mine next?" at a glance. The number uses the same order the scheduler dispatches in — #1 is what's coming out of the engine next. Record a memo, look at the pill, watch it tick down.

Custom titles show up

When you rename a recording inside Voice Memos — tap "Edit" and call it "Holmbacken 5b 55" or "Pappa berättar om sin släkt" instead of "Recording 47" — that title now shows up as the lead line on the row in My Transcriber. Until v0.3 those custom labels weren't read at all; the row just showed the timestamp.

The custom title is pulled from Voice Memos's own database (plaintext, despite the column being called ZENCRYPTEDTITLE). Memos you renamed before installing v0.3 backfill via a one-shot migration on first launch — you don't need to re-rename anything. Apple's default "Recording N" placeholder is filtered out so untitled memos stay clean.

A live view of what's transcribing

The Processing tab used to be a list of counters: how many audio files pending, how many in flight, nothing about which one was actually running. v0.3 turns it into a live feed. Under "Now processing" you see the queue itself, file by file, plus the in-flight Whisper transcription text as it streams out of the model. Long recordings are no longer a black box.

Stop, restart, retry

Two voice-memo controls that should have always been there:

• Abort. A new stop button next to the live progress pill cancels an in-flight transcription. The Whisper worker is preempted, the asset is marked Cancelled in frontmatter so it doesn't immediately re-queue, and the next item in the queue starts. Useful when you accidentally queued a 3-hour file.
• Transcribe now / Re-transcribe / Retry. The row context menu can force a memo to the front of the queue. The button used to be a silent no-op for pending memos; now it preempts the running transcription and the memo genuinely jumps the queue. The label is contextual: "Transcribe now" for pending, "Re-transcribe" for done, "Retry transcription" for failed or cancelled.

Defaults matched to your Mac

Two weeks of benchmarking Whisper across Mac hardware (the full numbers are here) rewrote the defaults My Transcriber picks on fresh installs. 8 GB Macs get medium — the largest model that runs comfortably without swapping. 16 GB and up jumps straight to large-v3-turbo, the model that finishes a 1-hour memo in ~2 minutes on Apple Silicon and ~30 minutes on a recent Intel Mac with CPU transcription.

Settings → Audio transcription still exposes every Whisper variant from tiny to large-v3-turbo; the detected tier just gives a sensible starting point so users on capable hardware aren't accidentally stuck on a smaller model. Already-configured installs aren't touched — the default-picker only fires once on fresh installs.

MCP for Claude Code

My Transcriber now ships an MCP server so Claude Code can query your voice memo index directly — ask "find the memo where I talked about the Acme deal" or "summarise last week's recordings" and Claude reads the index instead of being shown the files one by one. The bundled transcribe skill is auto-installed into your data folder on first launch so Claude knows how to drive the index without prompting.

The Integrations tab has a new MCP status card that reads your data folder's .mcp.json and shows live registration status, with one-click Reveal in Finder and a copy-to-clipboard cd "<folder>" && claude action. New MCP tools in v0.3: preview_bug_report and submit_bug_report, so Claude Code can troubleshoot a problem and file it for you in one session, going through the same HMAC-signed path that Settings → Bug report uses.

Long memos no longer time out

The previous guard against stuck transcription was a flat 60-minute wall-clock deadline. A one-hour memo on an Intel Mac — CPU-only, ~1× real-time per the bench — would routinely trip it despite making steady progress. v0.3 replaces the wall-clock timeout with a 10-minute idle deadline that resets on every progress callback — fails fast on a transcription that's actually stuck, never on a slow-but-progressing one.

Two related stability fixes: concurrent transcription requests are now serialized so the single-context Whisper server stops returning "busy" errors when the Voice Memos watcher, drop-folder ingestion, and the Transcribe tab all fire at once. And the background service's health-check timeout was bumped so it stops falsely reporting "offline" when the machine is under heavy CPU load.

Power Mode (battery-aware workloads)

A new Settings panel lets you gate background work on AC vs battery. For most My Transcriber users this is invisible: voice transcription defaults to Always, the same way it always ran — recording a memo and waiting for a transcript shouldn't depend on whether you're plugged in. The controls are there if you ever want to flip transcription itself to AC-only on the road, but the out-of-box behaviour matches v0.2.

Smaller things

• The Transcribe tab no longer shows "No transcriptions yet" on cold launch. The initial fetch was firing before the index finished opening; it now refreshes once the index is ready instead of latching to the empty initial result.
• Fresh installs no longer brick at "0 items indexed". The background service auto-creates ~/My Transcriber/, attaches iCloud as a source, and starts indexing on first launch. The old "pick a data folder" step is gone.
• Cross-machine voice memo re-imports no longer create duplicates. If you've ever synced your transcripts to a second Mac and run My Transcriber there, you used to end up with parallel duplicate transcripts. The importer now checks for an existing hash before writing, and a one-shot migration cleans up the parallel copies prior installs left behind.
• No more "Photo Library" permission prompt on startup. My Transcriber was inadvertently touching PhotoKit's status-read API on launch and macOS took that as a cue to surface the Photos permission dialog — confusing on a voice-only app. Fixed.
• Force-CPU and GPU-only diagnostic toggles are now hidden by default. The arch-aware default already picks the right backend per machine; the toggles were a footgun for everyone who isn't actively diagnosing a backend issue.

What's next

Not well defined yet. The immediate priority is stabilising v0.3 in real use — the next chapter will take shape from feedback. Bug reports from Settings → Bug report (or via Claude Code through MCP) go straight into that loop.