What's new in Vocateca?

Long shows no longer run your Mac low on memory. Transcribing very long episodes — multi-hour podcasts — could previously crash the app on Macs with less memory. Vocateca now processes the audio in small pieces and uses a small, steady amount of memory no matter how long the episode is.

Vocateca now eases off when your Mac is under memory pressure: it pauses starting new transcriptions until there's room again, instead of pushing on and crashing.

Smarter crash handling. Quitting the app mid-transcription is no longer mistaken for a crash, and a genuinely problematic episode is set aside sooner with a clear note instead of retrying and crashing over and over.

Back-catalog fills now run to the end. Steadily transcribing a show's older episodes no longer stalls after the first batch — it keeps going until the whole show is done.

Adding a YouTube playlist (or several videos at once) opens the Library once, when the whole batch has finished transcribing — not once after every single episode.

Fixed episodes that had finished transcribing still showing as “Pending” in the Library. The status is correct now and updates live, whichever tab you're on.

Deleting a transcript now asks first, everywhere — the row menu, the toolbar button and the new drag-to-delete all show a clear confirmation of exactly what will happen. Selecting several and deleting now removes the whole selection, not just one.

New: drag onto “Recently deleted” to remove things — one or more episodes (just their transcripts), or a whole show or person (unsubscribed, with transcripts kept in Recently deleted for 30 days). Each case confirms what it does first.

Shows “By person” now matches “By show”: open a show's full Details / Edit again (double-click, right-click, or the Details button), the stats bar stays when you switch modes, and the toolbar sits in its own row above the list.

Organise your transcripts into folders. Create a folder in the Library and drag whole shows or single episodes into it — removing something from a folder only unlinks it; your shows, episodes and transcripts are never touched. Select several at once with a checkbox, Shift-range or the space bar, and drag the whole selection in at once, Finder-style.

A new “Recently imported” view at the top of the Library lists your last 100 transcripts, newest first — with the full set of actions (reveal, transcribe next, add to Up Next, Send to Notion, export, delete) working across all your shows at once.

Exporting is overhauled: choose the file format, get sensible automatic names, export a folder of separate files or one combined file, all from a tidy “Export…” menu that can also Send to Notion. The action bar is always there now — it acts on what you've selected, or the transcript you're viewing.

Smaller touches: a persistent “+” in the toolbar opens Add from any screen, and the By-show / By-person switch now sits inline next to the search field.

Fixes: the Library no longer shows “0 of N” transcripts for minutes when you switch shows — the count is instant now and the deeper scan finishes quietly in the background. And the Status tab no longer clips the sidebar or slides its list under the toolbar on a short window.

Instagram is here, experimental: connect a dedicated Instagram account and follow a creator, and Vocateca fetches their reels and posts, transcribes the audio, reads the text inside images, and makes the captions searchable in your Library. Setup is one guided flow with a clear consent step — use a separate throwaway account, never your personal one; automated fetching is against Instagram's terms and can get an account limited or banned.

Once an episode is transcribed, its downloaded audio file is now removed by default to save space. Files you imported yourself are kept, and a failed transcription keeps its audio so it can be retried. You can turn this off in Settings if you'd rather keep the audio.

The setup guide now spells out what Pro actually unlocks — background fetching, per-show auto-download, folder watch and a daily summary — instead of just showing a badge. Everything else stays free.

Your subscription list is now protected from being accidentally overwritten: a suspicious write that would drop most of your shows, or blank their artwork and thumbnails, is refused and the fuller version kept — so a glitch can't quietly wipe your Shows.

When something used to fail silently, you now hear about it: if the download queue can't open its database it says so and asks you to restart, a single show's “Refresh feed” now reports a failure instead of doing nothing, and an automatic Notion push that fails leaves a note in the Notification Center.

Starting Vocateca at login is more reliable and no longer re-prompts once it's set up.

When a subscription loses its feed, reconnecting it is now one guided flow. Vocateca remembers where the show came from, opens the Add tab with the right source already searched and the best match pre-selected — you just confirm, and every transcript you already have is kept.

Fixed YouTube subscriptions that wrongly showed “feed unavailable”: a saved channel feed is now read correctly, so a refresh finds new videos again.

The Library no longer freezes for minutes on the first launch of the day — it scans transcripts in the background and remembers what it already checked.

Pro is now recognised the moment the app opens, even offline, from the signed licence itself — Pro accounts no longer show as Free until a network check catches up.

Before you subscribe, the Add preview now shows a podcast’s or channel’s description, its key facts, and its latest episodes.

An Instagram account that lost its link can be re-linked by confirming the account — or put off until tomorrow — instead of the show being reported as broken.

After a successful reconnect Vocateca returns to the Shows tab and selects the show; if it couldn’t find a match, it stays in Add so you can pick a different one.

Smaller fixes: the “+ Add” toolbar button opens the Add tab, deleting a show clears its still-queued episodes, and the first-run setup steps no longer shift around as you move through them.

Adding anything is now one place: the new “Add” tab. It searches podcasts and YouTube — no API key, no account — takes any link, and takes files from your Mac. The old Add popup, the “Local Ingest” tab and the YouTube Explorer have all moved into it.

You can search YouTube for a channel by name and see the subscriber count next to each hit, which is the only thing that tells four near-identically named channels apart. Pick one and the preview shows its latest uploads before you subscribe.

Instagram can't be searched, and the tab now says so from the first character you type instead of letting you conclude it from an empty result list. Instagram accounts are still added with their link.

A Spotify link that leads nowhere now explains why — no open feed, and the audio is copy-protected — instead of offering a “Try again” that cannot help.

The transcript now scrolls along with the video and highlights the line being spoken. It stops following the moment you scroll yourself, and picks up again when you stop.

Shows and Creators are one tab with a switch: view your subscriptions by show, or by the person behind them.

Vocateca now warns you before the disk fills up — a quiet bar first, a dialog when it gets serious — and offers to clear out media it no longer needs. If what it can free isn't enough, it says so and points at macOS's own storage tools rather than handing you a button that wouldn't fix it.

The first-run setup no longer makes you wait for the transcription model. The tools still have to be there, but the model finishes downloading in the background while you get on with it.

A new setup step gathers the two standing decisions — start at login, and fetch new episodes automatically — instead of burying them.

The tutorial now covers Background and Power mode, and Settings explains what the switch actually changes.

Settings no longer drops you into the middle of a category when you switch to it.

The “Keyword Watchlist” tab is now simply “Watchlist” — in the sidebar and its own title — and its ⌘-number, which had quietly stopped working, is back.

Both views of the Shows tab now have a search field — filter your subscriptions by show or by the person behind them — and the Settings search is now as wide as the Library’s.

The sidebar is tidier and narrower: it sizes itself to exactly fit the longest tab name and its unread count — growing only if that count needs another digit — with Settings and Help stacked at the bottom and the old stats line gone.

The action buttons in the Add tab’s video preview moved up into the window toolbar, matching every other tab — and the Status and Queue toolbars no longer spill their controls into a broken overflow menu.

Speaker recognition and Instagram are now marked Experimental, so you know where the rough edges still are.

Fixed a bug where a long episode could make Vocateca consume more and more memory until macOS forced it to quit.

Podcasts are no longer transcribed into the wrong language: when the detected language disagrees with a show's setting, Vocateca now detects the language from the audio instead of forcing the setting.

A show whose language was set wrong by an old version is corrected automatically the next time its feed is checked.

Importing an audio or video file from your Mac works again — and an imported file is never deleted by media cleanup, since it's yours, not a download.

Transcripts keep their words: common words like “so”, “ich” or “eine” are no longer turned into names picked up from the show.

When you subscribe to a podcast, its language now comes from the feed — and when the feed doesn't say, Vocateca asks instead of guessing.

Fixed shows that never fetched their episodes: after a few failed attempts Vocateca would stop checking a feed for days, and nothing you pressed — subscribing again, Repair, “Try again” — could talk it out of it. Asking for a refresh now always checks the feed.

A show with no episodes no longer claims the feed is unavailable when it isn't.

Typing a handle like “@lexfridman” now searches for it instead of assuming you meant Instagram — the same handles exist on YouTube. Instagram accounts are added with their link.

The YouTube Explorer is now a full feature (no longer beta): paste a link to watch the video beside a timestamped, scroll-synced transcript — pick the caption language, copy in any format, save to your Library, or subscribe.

Cleaner YouTube auto-captions: rolling, duplicated caption lines are collapsed into clean sentences automatically.

New Chrome extension: send a YouTube video to Vocateca with one click from your browser (Settings → YouTube).

Invite friends by sharing your referral link from the macOS share menu.

ffmpeg is found reliably — a Homebrew install works even when the app is launched from Finder, and Vocateca can install its own copy (incl. ffprobe) with no setup.

Your library loads and scans without stalling on large collections; finished transcripts resolve much faster.

Storage safety: the media cap never exceeds half of your Mac's free disk, and the Status tab now shows your free space.

A tidier Status tab, a searchable “Check for updates” in ⌘K, and clearer guidance when the disk is nearly full (open Storage settings / empty the Trash).

New export formats — CSV and WebVTT subtitles — available everywhere you choose a format, including a default in Settings.

Experimental YouTube Explorer (turn it on in Settings → YouTube): paste a link to watch the video beside a timestamped, scroll-synced transcript — pick the caption language, copy in any format, add it to your Library, or subscribe to the channel.

YouTube transcripts on the command line and through the MCP server, so you can automate transcript extraction from your own scripts and AI assistants.

Transcribe many videos at once — paste a YouTube playlist or a CSV of links.

New shows now backfill newest-to-oldest by default, with a setting to choose the order.

Much faster YouTube transcript loading.

Download your logs as a file — full or anonymized — and report a problem with the log attached, right from the Logs view.

A paused show no longer blocks manually starting one of its episodes, and the queue no longer reports “finished” when nothing was actually transcribed.

Only one copy of Vocateca runs at a time — deep links open in the running app instead of launching a second one.

Steadier under load — quieter and more useful logs, lower memory during long transcriptions, and a smoother launch on large libraries.

Fixed a rare crash during transcription while correcting proper nouns.

What's New now shows everything added since your previous version — not just the latest release — so nothing is missed after skipping an update.

“Check for Updates” in the About window now checks for updates directly instead of opening the website.

The sign-in popup is clearer: sign in right there, or jump straight to the Free-vs-Pro comparison for everything else.

Fixed the version number shown in About and Settings — it now always reflects the version and build you're actually running.

Invite friends to Vocateca — share your link, your friend gets a free month of Pro, and you earn a free month per year for every friend who stays Pro.

Automatic updates are now yours to tune — choose daily or weekly checks and whether updates install on their own, right from Settings.

Updates never interrupt your work: a running queue finishes its current transcription first, then updates and resumes automatically where it left off.

Branded, native update prompts show what's new before you install — with Install now, Remind me later, and Skip this version.

Update checks now explicitly send no system profile data — the app stays 100% telemetry-free.

A friendlier first run: decide whether Vocateca starts at login (no surprise system prompt) and finish with a clear "You're all set".

A polished installer — a branded disk image guides you to drag Vocateca to Applications and open it from there.

A dedicated What's New view after every update, so you never miss a change — or a new setting worth a look.

Queue controls now reflect live state at a glance: Start, Pause, and Stop enable exactly when they apply, with a clearer activity indicator.

Open any transcript in any export format straight from your Library, and act on several at once with multi-select.

Settings refinements — type directly into any number field, plus smoother scrolling.

Complete native Mac rewrite — faster, lighter, and built from the ground up for macOS.

On-device transcription, so your audio never leaves your Mac — a fast default engine, with Whisper as fallback.

Automatic speaker detection shows who said what in every transcript.

Subscribe to podcasts and YouTube channels, or bulk-import your existing subscriptions with OPML.

Watch folders transcribe local audio and video files automatically as soon as they land.

Export transcripts as Markdown, SRT, plain text, HTML, or the open OKF format.

Automatic proper-noun correction keeps names and terms consistent across transcripts.

Automate your workflow with the command-line tool, an MCP server for AI assistants, webhooks, and Notion/Obsidian sync.

Invite friends — share your link, your friend gets a free month of Pro, and you earn a free month per year for every friend who stays Pro.

Free forever for manual transcription, with Pro adding automatic background transcription — now available in English and German, with a privacy dashboard and automatic updates.

A full "Vapor" visual redesign brought a macOS Liquid Glass look in light and dark, following Apple's Human Interface Guidelines.

The UI moved from PyQt6 to PySide6, clearing the way for an open-core release of the code.

Paragraphos Pro made its debut: a Pro account, gated Schedule and folder-watch automation, and an upgrade flow — with active clicks staying free.

Transcription got smarter: per-episode language auto-detection, auto-vocabulary prompts drawn from past transcripts, and optional low-confidence word marking.

Reliability improved with automatic retries for transient failures, pre-transcribe integrity checks, a disk guard, and time-boxed undo for destructive actions.

New ingestion features: YouTube playlists, back-catalogue date backfill, re-upload detection, OPML import, and bulk export to Markdown, JSON, HTML, or PDF.

A new per-show episode browser, an expanded command-line interface, webhooks, and a local automation API brought full parity between the app and the CLI.

A single background-load level (quiet / balanced / full) replaced the separate parallelism knobs for transcription.

Adding a show from the command line now requires a --backlog choice, so an automated add can no longer silently transcribe an entire back-catalogue.

The running app no longer overwrites external edits to watchlist.yaml — it detects the change and merges it instead.

A show discovered outside the normal add flow is now flagged for review, with a banner to choose how much history to fetch.

A new AGENTS.md file documents how AI assistants can operate Paragraphos unattended.

A missed daily feed check — because the Mac was asleep or the app was busy — now runs automatically the next time the app comes to the foreground.

Paragraphos now also checks for app updates when brought to the foreground, not just at startup, capped at once every 24 hours.

A new Settings toggle turns update checking off entirely.

The "update available" notification now fires once per release instead of repeatedly.

Podcast search and cover-art fetching now share the app's pooled connection, fixing inconsistent timeouts.

Fixed a bug where the queue could reclaim episodes added mid-run, pushing the progress counter past the total.

Fixed the Library tab silently dropping recently transcribed episodes because of a filename mismatch — 16 of a user's 50 most recent episodes were invisible before the fix.

The Library tab now refreshes automatically as transcripts finish, instead of only at launch.

Fixed the Homebrew install helper reporting success even when the download had actually failed.

In-app log timestamps now include the date, not just the time, so long sessions are easier to read.

Fixed a batch of episodes failing to transcribe because the app rebuilt the wrong filename after a feed refresh, before checking whether the audio was already on disk.

Fixed some podcast episodes silently failing to transcribe because their file was actually M4A/ALAC despite an .mp3 name; such files are now converted before transcription.

Fixed local file imports being rejected outright, and YouTube episodes added by pasting a URL being sent through the wrong download path.

A dedicated Local Transcript tab replaced the old drop card, with a big drop area, a folder picker, and a URL field.

Dropping a file anywhere on the main window now jumps straight to the Local Transcript tab and starts the import.

Podcast search-by-name now shows results as you type, and a single click prefills the show details.

File ingest and URL probing no longer freeze the interface while they work.

Fixed dark-mode contrast on the Local Transcript hint text and broken file links for paths with spaces or umlauts.

Universal ingest: any audio or video file can now be dropped on the app, pasted as a URL, watched via a folder, or bulk-imported from a directory.

Full command-line parity with the app (23 commands), including JSON output for scripting and AI assistants.

Feed failures are now categorized (DNS, timeout, forbidden, and more) with a specific recommendation and a one-click retry.

The parallel-transcribe setting now actually runs multiple transcriptions at once, instead of always running one.

Fixed whisper silently producing no transcript for some podcast episodes because the app couldn't find Homebrew's ffmpeg when launched from Applications.

Fixed a class of feed failures on networks that synthesize IPv6 addresses, which had been wrongly rejected as private.

A startup log line now records the version, macOS, hardware, and every tunable setting, to make problems easier to diagnose.

Queue and Shows tab toolbars were reorganized to the top, with clearer sorting and a persistent Library page listing every transcript on disk.

YouTube channels became first-class shows: subscribe to a channel and new videos transcribe automatically.

A captions-first transcript path fetches uploader-provided captions instantly, falling back to whisper only when none exist.

Each channel can override its transcript source: captions, always whisper, or auto-captions as a fallback.

yt-dlp installs itself on first YouTube use and keeps itself updated weekly.

A new Settings filter can hide YouTube entirely, including skipping the yt-dlp install.

A new setup guide after the first-run wizard asks where transcripts should go and whether you use Obsidian, with automatic vault detection.

Podcast search results now show cover art, author, episode count, and the newest episode — loaded lazily as you scroll.

Scrolling to the bottom of search results automatically loads more, up to iTunes' 200-item cap.

A new Settings group lets you choose which output formats are saved; Markdown is always kept, SRT is now optional.

Fixed the Apple Podcasts add path leaving a show without a slug, and fixed deleting a show removing its transcripts along with the audio.

Fixed the first-run wizard failing to find a working whisper install on non-standard Homebrew locations.

Fixed brew install commands failing inside the app because Homebrew's own location wasn't on its PATH.

The wizard now rejects unsupported Macs (Intel, or macOS older than Ventura) upfront instead of wasting time on a doomed install.

Dependency installs now show live progress and a Retry button if one fails, instead of getting stuck.

The Queue now shows live transcription progress as a percentage on the active episode.

The estimated finish time is now based on actual audio duration rather than a per-episode average.

Fixed the progress counter not updating during a full-channel check.

Numerous UI polish fixes: window size now persists, dark-mode dropdowns render correctly, and in-app Logs/About panels replace Finder detours.

Downloads now run in parallel across multiple worker threads, with a per-host cap.

Interrupted downloads resume instead of restarting from scratch.

Right-click an episode to run it next or run it now, jumping the queue.

A keyboard shortcut cheatsheet (press ? or Cmd+/) lists every shortcut.

Settings now shows the whisper model's health — file size, hash, and whether it's drifted from the pinned version — with a one-click re-transcribe-all when it has.

Proper dark mode support: the app now follows the system appearance across every custom-painted element, with the accent color switching from ochre in light mode to purple in dark mode.

A new app icon and a menu-bar tray icon that auto-tints to match the menu bar.

The first release of Paragraphos, the Python/Qt app Vocateca grew out of.

A redesigned sidebar navigation replaced the old tabs, with live counts for Shows, Queue, and Failed.

A new Queue dashboard shows overall progress with a human-friendly finish estimate like "before lunch" or "tomorrow morning".

The Failed tab now explains errors in plain language, with actions to retry, mark resolved, or skip an episode forever.

RSS feeds refresh concurrently and skip unchanged feeds, and downloading plus transcribing now run as a two-stage pipeline.

Re-transcribing a single episode is now one click, keeping the old transcript around for comparison.

Bulk actions in the Shows tab let you disable, enable, mark stale, or delete multiple shows at once.

A three-mode Add Podcast dialog supports adding by name, by URL, or by pasting an Apple Podcasts link.

The very first build — a menu-bar app with a watchlist and a daily monitor that picked up new episodes on its own.

Library dedup by GUID and filename, and a filename sanitizer that kept umlauts intact.

An MP3 retention policy, a backlog filter, an RSS health check, OPML import, and a spot-check notification.

Verified against 16 real-estate podcast feeds on the first run: 2,023 reference episodes, nothing missed.