New in Vocello 3.0

Command line and agents.

vocello brings the Mac app's features to the terminal: generate in all three modes, long-form and line by line included; manage saved voices and models; read and export History; read and change settings. It uses the app's models, saved voices and settings, so both see the same library.

  1. Install the package

    Vocello-macos26-cli.pkg installs vocello in /usr/local/lib/vocello and links /usr/local/bin/vocello and its zsh completion. It is signed with Developer ID and notarized, asks for an administrator password, and runs on Apple Silicon (arm64) Macs with macOS 26 or later.

    vocello --help

    The installer adds links only in folders that root alone can change. If your /usr/local/bin belongs to your user, as some older Homebrew setups leave it, the command is installed but not linked there: run /usr/local/lib/vocello/vocello or link it into your path yourself.

  2. Or copy the folder

    Open Vocello-macos26-cli.dmg and copy the complete Vocello CLI folder to a folder you own, keeping its executable and resource bundles together, then run vocello by its full path. No administrator password is needed.

  3. Update

    Vocello > Check for Updates updates only the app. To update the command line, install the newer Vocello-macos26-cli.pkg from the releases page: it replaces the previous install. For a copied folder, replace it with the newer Vocello CLI folder.

  4. Uninstall

    The package's uninstaller removes the command line and leaves your models, voices and History untouched.

    sudo sh /usr/local/lib/vocello/uninstall.sh

    If users other than root can change /usr/local/lib, the uninstaller removes nothing: it prints the commands that make that folder root's alone. Run them, then run the uninstaller again. If a tool you installed since then needs that folder to stay yours, remove /usr/local/lib/vocello, /usr/local/bin/vocello and /usr/local/share/zsh/site-functions/_vocello yourself, then run sudo pkgutil --forget com.qwenvoice.cli.

First commands

Model files are not bundled. Install one once, in the app or from the terminal, and both use it. These commands list the speakers and models, download the Built-in Voice Speed model (about 2 GB, once), and save a first take to your Desktop.

vocello speakers listvocello models listvocello models install pro_custom_speedvocello generate --mode custom --variant speed --text "The train left at dawn." --out ~/Desktop/hello.wav

Voice Design and Voice Cloning each use their own model. vocello models install-recommended --plan lists the model each mode uses and its download size, and downloads nothing.

Built for scripts

--json and --ndjson give every command a stable machine contract: one envelope, typed error codes, documented exit statuses and progress events. Text output is for people and is not a contract. These commands pipe a script in for the JSON envelope, describe every command, option, error code and exit status, and send a JSON request on standard input; both takes use the Built-in Voice Speed model installed above.

echo "Hello there." | vocello generate --variant speed --text - --jsonvocello capabilities --jsonecho '{"mode":"custom","variant":"speed","text":"Good evening."}' | vocello run --request - --json
Some of the documented exit statuses
ExitCategoryMeaning
0okSuccess.
2usageThe invocation or its input was refused before any work.
69preconditionA model is not installed, or speech recognition is unavailable.
75transientThe engine or a store is busy, memory is short, or the network is unavailable: retry later.
77permissionVoice-cloning consent is missing, or a path is not allowed.
130cancelledInterrupted by Control-C (SIGINT).

For AI agents

vocello mcp serves Claude Code, Claude Desktop and other MCP clients over stdio. An agent can:

  • List voices, models, speakers and languages
  • Generate speech, with progress and cancellation
  • Enroll and delete saved voices
  • Read History and export takes

This adds Vocello to Claude Code for every project:

claude mcp add --scope user vocello -- /usr/local/bin/vocello mcp

The command names the installer package's link. With a copied folder, or when the installer added no link, use the path to your vocello executable instead: run mcp --print-config claude-code from that copy, for example /usr/local/lib/vocello/vocello mcp --print-config claude-code, and it prints the command with that copy's own path. --print-config claude-desktop prints the Claude Desktop setup the same way.

Limits to know

Not in History yet
Command-line takes are saved as files and do not appear in History yet; a 3.0 update adds that. Reading History from the command line needs the Mac app closed for now.
Consent every time
Cloning or saving a voice needs --confirm-consent on each command, or confirmConsent in each MCP request. It is never inherited from the app's Settings.
Files stay in Vocello
The MCP server reads and writes audio only inside Vocello's output folder, unless you start it with --allow-dir and name the folders it may use.
One engine on 8 GB
On an 8 GB Mac, one engine runs at a time and a running Vocello app counts: a second is refused as busy (exit 75), and --wait-for-engine waits for the other to exit.

Local by design. Yours to keep.

Vocello 3.0.0 for macOS 26 and Apple Silicon, free and open source. Three ways in, one local engine.

  • Mac app

    Open the DMG and drag Vocello to Applications. From 3.0 on, the app updates itself: choose Vocello > Check for Updates.

    Vocello-macos26.dmg · Signed and notarized

    Download for Mac
  • Command line

    The installer package links vocello in /usr/local/bin and asks for an administrator password; install a newer package to update it. Or copy the Vocello CLI folder from the folder DMG.

    Vocello-macos26-cli.pkg · Apple Silicon only

    Download the package
  • iPhone

    The public TestFlight beta runs version 2.4. The iPhone app's 3.0 version, on the same engine, follows in the coming days through TestFlight and then the App Store.

    iPhone 15 Pro or newer · iOS 26.0 or later

    Join on TestFlight

Models are not bundled: each downloads once from Hugging Face, so installing one needs an internet connection. Check the model sizes and your free storage first. The three Speed models take about 4.8 GB together, because they share one speech tokenizer, stored once. Generation then runs locally.

macOS 26.0+Apple Silicon with 8 GB memory or more (16 GB recommended for Quality models)Release notesAll releasesStable build for macOS 15: QwenVoice 1.2.3

View the source on GitHub