whisper-macos-cli 0.1.2

Transcribe audio files locally on Apple Silicon via whisper.cpp with Metal GPU acceleration, exposing a strict stdin/stdout JSON contract for AI agents and Unix pipelines.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95

# Privacy Policy

## Summary

whisper-macos-cli is a LOCAL-ONLY CLI. It does not collect, transmit,
or store any personal data on remote servers.

## What Data Is Processed

The CLI processes the following data on the local machine only:

- Audio files provided as arguments or via stdin
- Transcription text generated by the local Whisper model
- Diagnostic metadata (model name, duration, language, processing
  time) included in the JSON output

## What Data Is Stored

- Downloaded Whisper models are stored in
  `~/Library/Application Support/whisper-macos-cli/models/`
- No audio recordings are retained after transcription completes
- No transcriptions are retained unless the user redirects stdout
  to a file

## Network Activity

The CLI connects to `huggingface.co` only to download Whisper
models on first use. Each request includes:

- A User-Agent string identifying whisper-macos-cli and its version
- The standard HTTPS request headers
- The target URL of the model file

No cookies, no tracking pixels, no analytics, no telemetry.

## Subprocess and Temporary Files (v0.1.2+)

Since v0.1.2, the CLI may invoke `ffmpeg` as a subprocess to
extract audio from video containers. The following behaviors apply:

- The subprocess is invoked with `env_clear()` plus an explicit
  allowlist of `PATH`, `HOME`, `TMPDIR`, `LANG`, `LC_ALL`. The
  child process does not receive any other environment variables
  from the parent.
- A temporary WAV file is created in the system temp directory
  (resolved from `TMPDIR` or `std::env::temp_dir()`). The temp
  file is removed via a `Drop` guard even if the process is
  interrupted or panics.
- The temp file contains uncompressed PCM audio data derived
  from the user-supplied input. The audio content is the same
  as the input; the file is not transmitted to any external
  service.
- ffmpeg is an external binary, not bundled with whisper-macos-cli.
  Behavior depends on the user-installed version of ffmpeg.

## What Is NOT Collected

- No usage analytics
- No crash reports
- No personally identifying information
- No audio content
- No transcription content
- No IP addresses are stored or logged

## User Control

You can verify network behavior using the `doctor` subcommand, which
probes connectivity to Hugging Face and reports the result.

You can run the CLI in air-gapped mode by pre-downloading models
and using `--offline` to skip network checks.

## Model Weights

The Whisper model weights are downloaded from
`huggingface.co/ggerganov/whisper.cpp`. These weights are licensed by
their original authors (OpenAI). The CLI does not modify or repackage
the weights.

## GDPR and CCPA

This CLI does not process personal data of EU or California residents
because it does not transmit any data to remote servers. If you
transcribe audio containing personal data, you are responsible for
the lawful basis of processing under applicable law.

## Contact

For privacy-related questions, open an issue at
https://github.com/daniloaguiarbr/whisper-macos-cli/issues.

## Changes to This Policy

Changes to this policy will be tracked in `CHANGELOG.md` and announced
in release notes.