What you need to run tondi
Checked against v1.1.5, 12 September 2026.
tondi runs on a Mac with Apple Silicon (M1 or later) and macOS 13 or later. There is no Intel build and no Windows version yet; after a one-time 460 MB download it works without an internet connection.
Everything below is stated plainly, including what tondi can't do, so you can tell before downloading whether it fits how you work.
The requirements
| Computer | A Mac with Apple Silicon (M1 or later). Intel Macs are not supported. |
| macOS | macOS 13 or later. |
| Windows | Not available yet — get notified when it is. |
| Download size | 111 MB. Once installed, tondi takes about 284 MB on disk. |
| Security | Signed and notarized by Apple, under Developer ID team DHN5QUUM7B. macOS opens it without an unidentified-developer warning. |
| First run | Downloads a speech model of about 460 MB. This happens once. |
| Internet | Needed for the first-run model download, and to sign in or buy Pro. After that tondi works offline. |
| Speed | A ~10-minute video takes a few minutes to process. A 58-minute episode took 6 minutes 25 seconds on an M3 MacBook Air at Standard (measured 12 September 2026). |
| Input | .mp4, .mov and most common video formats. |
| Output | H.264 .mp4 clips, in vertical 9:16, square 1:1, portrait 4:5 or horizontal 16:9 — several shapes from one run. |
| Languages | Detected automatically; captions are in the spoken language. English is most accurate. |
| Highlight picking | A scoring system that weighs hook words, emphasis, numbers, speech rate and audio energy — not a language model. How it picks. |
| Account | None needed for Free. Pro signs in with a code emailed to you; there is no password. |
What tondi downloads, and when
The app itself is the 111 MB download. On first run it fetches a speech-recognition model from Hugging Face — about 460 MB for the Standard accuracy tier, which is the default. If you switch tiers, each one downloads its own model once: Fast is about 145 MB, Best about 1.5 GB. Models are stored on your Mac and never fetched again.
That is the only thing tondi downloads to make clips. Your video is never uploaded — transcription, highlight picking and rendering all happen on your Mac. The privacy policy lists every connection the app makes.
Common questions
Does tondi run on an Intel Mac?
No. tondi is built for Apple Silicon only — an M1 chip or later. There is no Intel build, and an Intel Mac will not be able to open it.
Is there a Windows version?
Not yet. A Windows version is coming; the homepage has a one-field form that emails you when it is ready.
Does tondi work offline?
Yes, once it is set up. The first run downloads a speech model of about 460 MB; after that, making clips needs no internet connection. Signing in and buying Pro need a connection, and a Pro subscription keeps working for seven days offline between checks.
How long does a video take to process?
A roughly 10-minute video takes a few minutes at the Standard accuracy tier; a 58-minute episode took 6 minutes 25 seconds on an M3 MacBook Air. Longer videos take proportionally longer; Best is slower and Fast is quicker.
What video files does tondi accept?
Anything ffmpeg can read, which covers .mp4, .mov and most common formats. Clips are exported as H.264 .mp4 files in the aspect ratios you choose.
What languages does tondi transcribe?
tondi detects the spoken language automatically and captions in that language; it does not translate. English gets the most accurate captions. Other languages work, and the Best tier helps when Standard makes mistakes.
How much disk space does tondi need?
The download is 111 MB and the installed app takes about 284 MB. Add the speech model — about 460 MB for Standard, 145 MB for Fast, 1.5 GB for Best, each downloaded once — plus the clips you make.
Fits? Try it on your own footage
Free, no account, no card. Three clips per video with a small watermark; Pro lifts the limits for $10 a month. Clipping client episodes? tondi for podcast editors.
Download for macOSv1.1.5 · 111 MB · macOS 13+ · Apple Silicon