Scribiz changelog
What changed in Scribiz, by date, newest first. Each line says what you can do now.
Scribiz updates
Mac app 0.1.11: call audio with earbuds
- FixedRecording a call on your Mac now includes the other side of the conversation even when you wear earbuds or headphones. System audio is captured alongside your microphone and mixed into the recording automatically.
Mac app 0.1.10: free on your Mac, and a calmer design
- BetterA first run anyone can follow. The Mac app asks how you want to start: sign in, free on this Mac (one download and you are ready), or your own Gemini key.
- FixedOpening a recording in the Mac app works again. In 0.1.7 the whole window could turn blank on recent macOS when you clicked a recording.
- NewFree, offline transcripts on your Mac now use NVIDIA Parakeet. Pick Parakeet v3 in Models and download it once, nothing else to install. It reads 30 minutes in about 12 seconds with a third fewer mistakes than the Whisper model it replaces.
- BetterA calmer Mac app: small page titles, settings as a plain list, Models, Output, Account and Settings in the sidebar, and more room for the transcript when you open a recording.
- NewThe Mac app tells you when a newer version is on the download page, at the bottom of the sidebar and under Scribiz, Check for Updates.
Mac app 0.1.7: voice notes, Ask anything and History sync
- NewMac app 0.1.7 has a new notes-style design. Home and History show each title with a two-line overview and a play button, and results open as a single document with Summary, Transcript and On screen tabs.
- NewRecord a voice note from anywhere in the Mac app. Press the red Record button at the bottom of the window, press ⇧⌘R, or start from the menu bar. Recording keeps going in the background even when the window is closed, then transcribes and summarizes the note automatically.
- NewAsk any result a question in the Mac app. Type into Ask anything at the bottom of a result to get an answer with clickable timestamps.
- BetterSigned-in Mac app, CLI and MCP runs now sync their finished transcripts and summaries to your Scribiz account History so they appear across all your devices.
Pause removal in noisy recordings
- BetterPause removal works on noisy recordings. On a result page, Long ones and Every one cut almost nothing in a recording made on a plane, in a car or with a phone recorder, because the steady noise looked like speech. They now find those pauses: 7 to 9 cuts in a one-minute voice memo recorded on a plane, where there were one or two. Clean recordings are cut exactly as before.
- FixedSpeech skipped in a noisy recording is read again. Scribiz checks every transcript against the speech it hears in the audio and reads again what the speech model left out. In steady noise that check heard too little speech, and it missed the first 16 seconds of a voice memo recorded on a plane. It now notices and reads that part again.
- BetterParagraphs end at the end of a sentence. The transcript view and the text and Markdown downloads no longer break a paragraph in the middle of a sentence at a short pause or after 120 words.
- BetterThe full proofread fixes names in short lines. A safety check meant to stop it moving words between lines also refused fixes like "Spring Studio" to "Screen Studio" and "double funnel" to "top-of-funnel". They go through now, and its message says what it means: "The proofread left 5 of 28 lines as heard".
- FixedThe full proofread no longer prints a word twice where two lines meet. When it moved a word from one line into the next ("open Screen Studio"), the word could end up in both. It now puts the word back in its own line and keeps the fix.
- BetterShort two-word clauses no longer flash by as their own subtitle. A one or two word phrase before a comma ("Screen Studio,") now stays with the words that follow in SRT, VTT and the player captions instead of appearing for half a second on its own.
- FixedPhone calls with long holds keep their speaker turns and stay quiet during hold music. Hold music, keypad beeps and ringing no longer trigger false missing-speech repairs or warnings, short one- and two-word replies ("Yeah.", "Okay.", "Yes.") stay on their own speaker instead of merging into the other speaker, and speaker labels stay consistent after a long hold.
- NoteThe command-line tool 0.1.7 (on npm) and Mac app 0.1.7 (on the download page) include these changes.
A new speech model, and a picker on the way
- BetterA new speech model reads your audio. Files you upload, links whose audio we download and voice notes you send in Telegram are now transcribed by ElevenLabs Scribe v2. In our measurements it got more words right on short and quiet recordings, reads a long recording in about half the time, and keeps speaker labels and word times. Minutes are counted the same way as before.
- NoteNext: choose your own speech model. Scribiz stays the default, and the command-line tool and the Mac app will also take your own Gemini or OpenRouter key, or run Whisper on your computer, free and offline, with no account. It is in the command-line tool 0.1.5 and in the Mac app 0.1.5, both out now. The new Speech models page in the docs explains it.
Bulk YouTube transcripts and separate caption credits
- NewGet YouTube transcripts in bulk from video links, playlists or channels. Select the videos, preview the credit cost and read results as they finish. Cancel a batch, retry failed items or resume it later.
- NewDownload a ZIP or one combined Markdown or TXT file with video titles, source links and your selected order. CSV and JSON manifests include the status of each video. Available source timestamps stay attached to the captions.
- NewYouTube caption credits are separate from Listen and Watch minutes. An account gets 30 free caption credits once. One-time packs cost $3.99 for 200, $11.99 for 1,200 or $23 for 5,000 credits. Credits never expire. A new successful transcript costs one credit regardless of length. Failed items, stored reads and exports use no credits.
- BetterTranscript-only YouTube requests use existing caption tracks without downloading video or audio or running AI. If a track cannot be delivered, that item reports the problem. Cleanup, summaries and video understanding remain separate actions.
- NewThe same caption workflow is available through the API, MCP and the scribiz 0.1.3 command-line tool. The signed and notarized Mac 0.1.4 installer is on the download page. A separate YouTube transcript MCP page explains setup for Claude Code, Cursor and Codex.
- NoteA settled caption-pack purchase raises the daily retrieval allowance. If retrieval pauses at the daily limit, resume unfinished items after midnight UTC. Stored results stay available.
- NewHLS links work in the API and the MCP server: pass the address of a playlist that ends in .m3u8. Live streams and encrypted streams are not read.
- BetterThe command-line tool, scribiz 0.1.4, checks your minutes before a run when you are signed in. The plan is quoted in minutes beside what you have left, and a video longer than your minutes is refused before anything is spent. If the minutes run out at the proofread, the transcript is still written, with a warning. The closing line shows what the run used and what is left.
- NoteFrom now on a full proofread asked for through the API uses half a minute more for each minute of video, so Listen with it uses 1.5. Proofread "full" and a vocabulary both count. The quote shows it before the run, and it is charged only if the proofread ran. It needs an account. The remote MCP tools do not run the full proofread.
The command-line tool is on npm, with sign-in
- NewThe command-line tool is on npm as scribiz. Install it with npm install -g scribiz, or run npx scribiz. It runs on your machine, reads files, folders and links, and writes SRT, VTT, TXT, Markdown, Context or JSON. It needs Node 24 and ffmpeg, plus yt-dlp for most links, and has been tested on macOS only.
- NewVersion 0.1.1 adds command-line sign-in. Run scribiz login, confirm the code it shows at scribiz.com/login/device, and a Scribiz key is saved on your machine. A free account has 30 minutes a month, scribiz whoami shows how many are left, and when they are used up the tool says when they come back. Your audio goes from your machine to Scribiz, which sends it to Google.
- NewOr skip the account: scribiz setup saves your own Gemini API key, and then your audio goes from your machine straight to Google. One credential is saved at a time, so login and setup replace each other, and GEMINI_API_KEY in the environment wins over a saved sign-in. scribiz logout removes the saved credential from your machine and does not revoke the Scribiz key: the dashboard does.
- NewA local MCP server comes with it: claude mcp add scribiz -- scribiz mcp. It runs on your machine with your Scribiz account or your own Gemini key. In 0.1.1 it takes the watch argument on get_video_context and ask_video and serves on-screen notes that are already stored, like the hosted server with a key. It refuses links to your own machine and to private networks unless you pass --allow-private-network.
- NewThe Mac app is out, for Macs with Apple silicon on macOS 14 or newer: drop in a video, audio file, folder or link and read the transcript, the text on screen, the summary and the chapters, then export SRT, VTT, text, Markdown or JSON. Download it from the download page; it is signed and notarized by Apple.
Scribiz opens
- NewThe free web tool. Paste a video link or drop an audio or video file and get the transcript, what was on screen, a summary and chapters. No account is needed to start.
- NewExport as SRT, VTT, TXT, Markdown, Context or JSON. Speakers are labelled when there is more than one voice.
- NewFour modes: Auto, Listen, Watch and Both. Auto picks the cheapest path that works. Watch reads the text on screen, so a silent screen recording still gets a result.
- NewA remote MCP server at https://scribiz.com/mcp for Claude, Cursor, Codex and other agents. It needs no key. It has five tools: get_transcript, get_video_context, ask_video, search_video and get_job.
- BetterThe MCP server now reads videos that have no captions our server can read. Without a key it uses captions when it can, serves videos Scribiz has already processed at no cost (a summary it has not written for one yet uses a tenth of the video length), and otherwise a model reads the link, inside the same 10 minutes a day as the free web tool (videos up to 15 minutes, 5 questions a day). Times are approximate when a model read the link, and it never looks at the picture, so there are no on-screen notes. Everyone who connects without a key also shares one daily limit. When it is used up, captions and videos already processed still work.
- NewA REST API at https://scribiz.com/api. Create a job from a link or an upload, follow it, read the result and ask a question about it.
- NewDocs for the web tool, the MCP server and the API at /docs.
- BetterA video that Scribiz has already processed is served again in the mode you asked for, and costs no minutes.
- NewTwo more tool pages: M4A to text for voice memos and call recordings, and a YouTube subtitle downloader for SRT and VTT.
- NewGuides at /blog: how to get a YouTube transcript when there are no captions, how to open the transcript on YouTube, and whether ChatGPT or Claude can watch a YouTube video.
- NewComparisons at /alternatives: Scribiz next to Tactiq and NoteGPT, with what each tool does better.
- NewAbout and Changelog pages.
- BetterThe docs now say plainly that the command-line package and the Mac app are not published yet, and show what works today: the web tool, the MCP server and the API.
- BetterThe privacy policy now says what the site does about analytics, and the terms and the docs state the captions rule one way: a tenth of a minute for each minute of video, and 30 lookups a day without an account.
- NotePlans are Free, Pro ($10 a month or $84 a year) and Mac Lifetime ($99 once, $59 early bird for the first 500). Pro and Mac Lifetime are not on sale yet, and the Mac app cannot be downloaded yet.
Before Scribiz: the transcribe tool
Scribiz follows transcribe, an open-source command-line tool by the same maker. These are its releases. transcribe on GitHub.
transcribe 4.2
- BetterOne-step setup: paste a key and the tool works out which service it belongs to. A doctor command shows which keys and tools are in place. The clean-up pass uploads far less audio and runs faster.
transcribe 4.0
- NewSpeaker labels, an optional second pass that fixes mis-heard names and technical terms while keeping the timing, and support for other services that speak the same API, such as OpenRouter.
transcribe 3.8
- NewA right-click action in Finder on a Mac: Transcribe Subtitles.
transcribe 3.7
- NewInstagram Reels, using your browser login.
transcribe 3.5
- NewPass a folder and pick the files to transcribe in bulk.
transcribe 3.4
- NewScreen Studio recordings.
transcribe 3.3
- NewLinks from other sites that yt-dlp supports, such as X, not only YouTube.
transcribe 3.2
- BetterLong files are transcribed in parallel pieces, so they finish sooner.
transcribe 3.1
- NewLong files are split automatically. New options for where to save the file and for shifting timestamps in an editor.
transcribe 1.0 and 2.0
- NewFirst release: one command turns an audio or video file into SRT subtitles. YouTube links worked the same day.
Got a video? Get the transcript.
Or