Argmax Transcription and TTS

v1.0.1

On-device speech-to-text (Whisper) + text-to-speech (Qwen3-TTS) CLI. Runs on the Apple Neural Engine (ANE), Apple's low power, dedicated ML inference chip. M...

0· 258·0 current·0 all-time
byZach Nagengast@zachnagengast
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
VirusTotalVirusTotal
Benign
View report →
OpenClawOpenClaw
Benign
high confidence
Purpose & Capability
Name/description (on-device STT + TTS) aligns with declared requirements: a single binary (whisperkit-cli) and a Homebrew install. No unrelated credentials, binaries, or config paths are requested.
Instruction Scope
Runtime instructions are scoped to saving incoming audio to /tmp, invoking whisperkit-cli for transcribe/tts, and attaching generated files. SKILL.md documents a one-time model download from HuggingFace on first run and a local OpenAI-compatible server (binds to 127.0.0.1 in examples). These behaviors are reasonable for the stated purpose but imply network activity only at setup (model download) and a local HTTP endpoint that other local processes could call.
Install Mechanism
Install is via a Homebrew formula (whisperkit-cli). Using Homebrew is a standard, low-risk distribution method compared with arbitrary downloads; the SKILL.md points to a GitHub project. Users should still confirm the formula source (tap) is trustworthy before installing.
Credentials
The skill requests no environment variables, credentials, or unrelated config paths. This is proportionate to a local CLI tool that performs on-device inference.
Persistence & Privilege
The skill does not request elevated platform privileges or always:true. It will install a binary via Homebrew and download/keep model files on disk (potentially hundreds of MBs to GBs). It also documents running a local server; ensure you understand where models are stored and that the server binds to localhost only.
Assessment
This skill is internally consistent with its description: it installs a Homebrew CLI that runs locally and requires no API keys. Before installing, verify the Homebrew formula/tap and the GitHub repo (https://github.com/argmaxinc/WhisperKit) are legitimate. Be aware that on first run the tool will download models from HuggingFace (large files, network activity) and will store them locally—check disk usage and model storage location. The skill can also start a local HTTP server (examples show 127.0.0.1:50060); ensure it’s bound to localhost and not exposed to external networks. If you need stronger assurance, inspect the brew formula source or run the CLI in a controlled environment (VM or dedicated machine) before adding it to an agent that processes sensitive audio.

Like a lobster shell, security has layers — review code before you run it.

latestvk97f6creamb4n64gzy0c06jqy1826373

License

MIT-0
Free to use, modify, and redistribute. No attribution required.

Runtime requirements

🗣 Clawdis
OSmacOS
Binswhisperkit-cli

Install

Homebrew
Bins: whisperkit-cli
brew install whisperkit-cli

Comments