Skip to content

Releases: Desert-Ant-Labs/desert-ant-cli

desertant 0.5.0

Choose a tag to compare

@github-actions github-actions released this 20 Oct 19:59

desertant 0.5.0 follows desert-ant-core 4.1.0.

  • To run several models on a file in one command, run a flow: da flow run podcast-clips episode.mp4 cleans up the sound, transcribes it, and cuts captioned clips. da flow templates lists nine ready-made flows, and da docs flows explains how to write your own.
  • da clips --captions burn draws the captions into each clip, in white Inclusive Sans over a soft shadow, so a clip looks the same on macOS and Linux. A vertical clip gets one short line at a time, and a square or wide clip up to two lines.
  • --caption-style shadow|outline|box, --caption-position, --caption-align, --caption-font, --caption-size, --caption-weight, --caption-color, --caption-lines, and --caption-words set how burned captions look.
  • --quality high, the default, burns captions at the highest quality, and medium and low make smaller files. Every level keeps the clip's size and frame rate.
  • da clips --captions srt writes an .srt beside each clip.
  • da schema --json lists every flow step, with its inputs, outputs, and settings.
  • On Linux without ffmpeg, da clips says it needs ffmpeg before it transcribes, instead of after.
  • In a pipe or a file, stdout now holds the result alone, so da voz talk.mp4 --format srt > talk.srt writes a valid .srt. Notes and next-step hints show only on a terminal.
  • On macOS, --force replaces an existing clip without re-encoding it.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.4.0

Choose a tag to compare

@github-actions github-actions released this 19 Oct 19:24

desertant 0.4.0 follows desert-ant-core 4.1.0.

  • da clips runs on Linux. Voz transcribes the recording, Clips picks the moments, and ffmpeg cuts the clips.
  • On macOS 27, da uhm runs on Core AI, with new weights, v1.2.0.
  • On an Intel Mac, the install script says that desertant needs Apple silicon. In a Rosetta shell on Apple silicon, it installs the Apple silicon build.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.3.0

Choose a tag to compare

@github-actions github-actions released this 19 Oct 16:35

desertant 0.3.0 follows desert-ant-core 4.0.0.

  • To label who said what in a recording, run da who talk.mp4. To transcribe and label in one go, run da voz talk.mp4 --speakers --names "Anna,Ben". The SRT, VTT, and text files carry the names.
  • To find photos in a folder, run da eye photos/ --has beach --not blurry. --duplicates groups near-duplicates with the keeper first, --best picks one photo, --sort aesthetic ranks them, and --crop 1:1 writes a crop. Eye is in closed beta, so set HF_TOKEN to a Hugging Face token with access to desert-ant-labs/eye.
  • To pick a video's thumbnail, run da eye launch.mp4 --thumbnail. --has people lists the stretches where people are in frame.
  • desertant runs on Linux, x86_64 and arm64, with glibc 2.35 or later. The install script works there too. ffmpeg reads compressed audio, video, and images, and libheif reads HEIC.
  • da voz transcribes on Linux.
  • da ear names the spoken language from 102 languages, up from 99.
  • On a Mac without a Neural Engine, such as a virtual Mac, da redact warns that it does not detect names, addresses, or cities there.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.2.2

Choose a tag to compare

@github-actions github-actions released this 11 Oct 18:37

desertant 0.2.2 follows desert-ant-core 3.5.0.

  • da tongue runs again. 0.2.1 left Tongue's weights out of the release and crashed with "unable to find bundle named DesertAnt_Tongue".
  • If you installed with mise, run mise use -g "ubi:Desert-Ant-Labs/desert-ant-cli[extract_all=true]" once. The old line installed only the binary, so Tongue and Title could not find their files.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.2.1

Choose a tag to compare

@github-actions github-actions released this 28 Sep 12:15

desertant 0.2.1 follows desert-ant-core 3.5.0.

  • Models load faster on macOS. On an M4 Max, da voz on a 3s clip finishes in 0.5s instead of 3s, and da gist in 0.16s instead of 0.56s.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.2.0

Choose a tag to compare

@github-actions github-actions released this 23 Sep 20:57

desertant 0.2.0 follows desert-ant-core 3.4.0 and adds Moderator and Tongue.

  • Check a photo for nudity or sexual content with da moderator photo.jpg. Moderator runs on Apple silicon.
  • Detect the language of a short text with da tongue "<text>". Tongue needs three words and knows 84 languages.
  • Transcribe every recording in a folder by passing the folder: da voz interviews/. Every command that reads files takes a folder or several paths, and --recursive includes the folders inside. The model loads once for the whole folder.
  • With several files, --json prints one array with a result per file, in the order they ran. If a file fails, the CLI reports it on stderr and keeps going. One file still prints a single result, so scripts that parse it keep working.
  • See where the CLI installs and caches its files, and how da update works for each install method, with da docs files.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.1.2

Choose a tag to compare

@github-actions github-actions released this 23 Sep 12:37

desertant 0.1.2 follows desert-ant-core 3.3.1.

  • da clips and the other audio commands read recordings with many channels and files of several gigabytes.
  • da models and da info show the 3.3.1 catalog: Align lists Linux and Windows, and Clear lists Core AI.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.1.1

Choose a tag to compare

@github-actions github-actions released this 03 Sep 11:09

desertant 0.1.1 follows desert-ant-core 3.1.0.

  • The loading label says a model is slower the first time only on its first load on a Mac, and again when its weights change.
  • da voz reports how long the model took to load apart from the transcription, and the realtime factor covers transcription alone.
  • Timestamps in da voz, da uhm, and da clips print in dark teal.

Install or update with da update, brew upgrade desertant, or mise upgrade.

desertant 0.1.0

Choose a tag to compare

@github-actions github-actions released this 03 Sep 10:14

desertant 0.1.0 follows desert-ant-core 3.1.0.

Run Desert Ant Labs models on your own machine from the terminal: transcribe a recording, cut it into clips, clean up the audio, redact personal data from a text, suggest emoji, or tag a topic. Weights download once. Nothing leaves the device.

The command is desertant; da is the short alias. Nine models run today: voz, clips, clear, uhm, ear, redact, emo, gist, and title (Apple silicon). Every command takes --json, reads stdin, and exits 0, 1, or 64, and da schema --json describes each command and how they chain. desertant setup writes the adapter for Claude Code, Codex, or Pi.

Install with the script, Homebrew, or mise; desertant update brings any of them to the next release. macOS on Apple silicon, and Linux on x86_64 and arm64.

curl -fsSL https://raw.githubusercontent.com/Desert-Ant-Labs/desert-ant-cli/main/install.sh | sh