dictate-mlx 2.5.4

Push-to-talk voice dictation that runs entirely on your Mac.
No cloud. No API keys. No subscriptions.

Hold a key → Speak → Release → Clean text appears wherever your cursor is.

Why Dictate?

• Local speech recognition accelerated by Apple Silicon
• ANE acceleration for STT — reduces GPU contention; model memory still comes from unified memory
• Local inference by default — optional remote text cleanup requires explicit opt-in
• Free and open source — no subscriptions, no API keys, no accounts
• LLM text cleanup — local model fixes grammar and punctuation automatically
• Multiple languages — engine coverage varies; optional LLM translation

Dictate can use your Mac’s Neural Engine for speech recognition.

Install

python3 -m venv ~/.venvs/dictate
source ~/.venvs/dictate/bin/activate
python -m pip install --upgrade dictate-mlx
dictate

That's it. Dictate launches in the background and appears in your menu bar. Close the terminal — it keeps running.

For Qwen3-ASR support (30 languages plus 22 Chinese dialects):

python -m pip install --upgrade "dictate-mlx[qwen3-asr]"

This is still local-only. No API key is required; the extra installs the MLX Qwen3-ASR runtime.

Homebrew source install is available for users who prefer Brew. It builds the Swift ANE helper and installs the Python app into a Homebrew-managed virtualenv:

brew tap 0xbrando/dictate https://github.com/0xbrando/dictate
brew install dictate

The cask/DMG path is still planned; pip install dictate-mlx remains the simplest install path.

macOS will prompt for Accessibility and Microphone permissions on first run. Dictate downloads only the selected default models, then caches them in ~/.cache/huggingface/. Other cleanup models are one-click downloads from the Quality menu.

git clone https://github.com/0xbrando/dictate.git
cd dictate
python3 -m venv .venv
source .venv/bin/activate
pip install -e .
dictate

Requirements

• macOS with Apple Silicon (any M-series chip)
• Python 3.11+
• Several GB of unified memory and disk space for the selected models

Features

Push-to-Talk

Hold a key, speak, release. Text appears wherever your cursor is.

The PTT key is configurable: Left Control, Right Control, Right Command, or either Option key.

LLM Text Cleanup

The thing that sets Dictate apart. Most dictation tools give you raw transcription. Dictate pipes through a local LLM that fixes grammar, adds punctuation, and formats properly.

Clean, already formatted phrases of up to 8 words can skip cleanup. Translation and other writing styles still use the LLM.

Local STT Engine Stack

Dictate is designed around local speech recognition. Switch anytime from the menu bar.

ANE is the default. It runs Parakeet through FluidAudio and Core ML. This reduces GPU contention, but Apple Silicon shares unified memory across processors. Dictate currently transcribes and then cleans text sequentially. Latency depends on clip length, model, hardware, and whether models are warm.

Qwen3-ASR is the recommended local multilingual engine — 30 languages and 22 Chinese dialects, including Japanese, Chinese, and Korean. Requires pip install dictate-mlx[qwen3-asr].

Dictate auto-switches engines based on language: ANE/Parakeet for European languages, Qwen3-ASR for CJK and others, Whisper as the universal fallback.

Writing Styles

Toggle LLM cleanup off from the menu bar for raw transcription output.

Real-Time Translation

Speak in one language, get output in another. 12 languages supported: English, Spanish, French, German, Italian, Portuguese, Japanese, Korean, Chinese, Russian, Arabic, Hindi.

Quality Presets

Dictate does not install every LLM up front. First run downloads the recommended default for your Mac; selecting another local Quality preset downloads that model once and reuses it forever. If you already run Ollama, LM Studio, vLLM, or another OpenAI-compatible localhost server, choose Local API Server to avoid loading a Dictate-managed cleanup model.

Clean phrases of up to 8 words can skip cleanup. The app picks a default model for your chip; compare presets on your own dictation.

Recommended defaults:

End-to-End Pipeline

Dictate records a clip, transcribes it, optionally cleans the transcript, then pastes it. Model download and first-load time are separate from warm inference time. Older 65ms figures are specific STT measurements, not an end-to-end latency guarantee.

See the September 2026 audit for measured results on an M2 Max and current model recommendations.

Menu Bar

Everything accessible from the waveform icon:

• Writing Style — Clean Up, Professional, Bullet Points
• Quality — Fast, Balanced, Quality, or localhost API server; missing models download when selected
• Input Device — select microphone
• Recent — last 10 transcriptions, click to re-paste
• STT Engine — ANE (default), Qwen3-ASR, Parakeet, or Whisper
• PTT Key — choose your push-to-talk modifier
• Languages — input and output language
• Sounds — 6 notification tones or silent
• Personal Dictionary — names, brands, technical terms always spelled correctly
• Launch at Login — auto-start on boot

ANE Engine Setup

The ANE (Apple Neural Engine) engine is the default and recommended STT engine. It requires a small Swift binary that Dictate calls behind the scenes. If the binary isn't installed, Dictate falls back to Parakeet (GPU-based STT).

# Build from source (requires Xcode command line tools)
cd swift-stt
swift build -c release

# The binary lands at swift-stt/.build/release/dictate-stt
# Either add it to your PATH or leave it — Dictate finds it automatically

First run: CoreML models download automatically (~2.7GB) and compile for your chip. This takes 1-2 minutes the first time. After that, models are cached and transcription starts instantly.

Requirements: macOS 14+ (Sonoma or later), Apple Silicon.

What it does: The dictate-stt binary uses FluidAudio to run Parakeet speech recognition on the Neural Engine via CoreML. All processing is local — no network calls after the initial model download.

dictate-stt check                    # Verify ANE is available
dictate-stt transcribe recording.wav # Transcribe a WAV file
dictate-stt serve                    # Keep models warm for repeated requests

Local API Server

If you run a local LLM server, Dictate can use it instead of loading its own model — zero additional RAM:

DICTATE_LLM_BACKEND=api DICTATE_LLM_API_URL=http://localhost:8005/v1/chat/completions dictate

Works with any OpenAI-compatible server on your Mac: vllm-mlx, LM Studio, Ollama.

The API Server preset is still local-first. Remote URLs are blocked unless you explicitly set DICTATE_ALLOW_REMOTE_API=1.

Cloud Policy

Dictate does not need cloud services. Audio and text stay on your Mac by default.

Cloud endpoints are intentionally opt-in only:

DICTATE_ALLOW_REMOTE_API=1 DICTATE_LLM_API_URL=https://example.com/v1/chat/completions dictate

Do not enable this unless you understand that cleaned-up text may leave your machine. Speech recognition remains local unless you replace Dictate's STT pipeline yourself.

Environment Variables

Agent Integration

Dictate works well as a voice input layer for AI assistants and agent frameworks. If you're building with tools like Claude Code, OpenClaw, or similar — Dictate gives your setup a local, private voice interface with zero cloud dependency.

CLI Commands

dictate              # Launch in menu bar (backgrounds automatically)
dictate config       # View all preferences
dictate config set writing_style professional
dictate config set quality fast
dictate config set ptt_key cmd_r
dictate config set stt whisper
dictate config reset # Reset to defaults
dictate stats        # Show usage statistics
dictate status       # System info and model status
dictate doctor       # Run diagnostic checks (troubleshooting)
dictate devices      # List audio input devices
dictate update       # Update to latest version
dictate -f           # Run in foreground (debug)
dictate -V           # Show version

Config Keys

Shell Completions

Tab completions for bash and zsh:

# Bash — add to ~/.bashrc
source /path/to/dictate/completions/dictate.bash

# Zsh — copy to fpath dir, then reload
cp completions/dictate.zsh ~/.zsh/completions/_dictate
autoload -Uz compinit && compinit

Completes commands, config keys, and all valid values.

Debugging

If recording finishes but no text appears, enable Dictate in System Settings → Privacy & Security → Accessibility, then quit and reopen Dictate. For a terminal launch, grant the terminal app instead. This permission lets Dictate send the paste shortcut. Click into an editable text field before holding the push-to-talk key. Failed output stays in Recent; if copying succeeded, you can also paste with ⌘V. Public model downloads do not require a Hugging Face account or Keychain access.

# Run in foreground with logs
dictate --foreground

# Check background logs
tail -f ~/Library/Logs/Dictate/dictate.log

Security

• Audio inference is local. Optional remote text cleanup requires explicit opt-in. Output is copied to the system clipboard before pasting.
• Temporary audio files stored in a private directory with owner-only permissions — not world-readable /tmp.
• The ANE engine's dictate-stt binary is open source Swift code you build yourself from swift-stt/. CoreML models download from Hugging Face on first run, then everything is cached locally.
• Models restricted to the mlx-community/ HuggingFace namespace only.
• LLM endpoints restricted to localhost by default (DICTATE_ALLOW_REMOTE_API=1 to override).
• Preferences and stats stored with 0o600 permissions (owner-only read/write).
• Log rotation (5MB, 3 backups) prevents disk exhaustion.
• HuggingFace telemetry disabled at startup (DO_NOT_TRACK=1).
• No API keys, tokens, or accounts required. No unsafe code patterns.

Contributing

Issues and PRs welcome. Run the test suite before submitting:

python -m pytest tests/ -q

See CONTRIBUTING.md for guidelines.

License

MIT — See LICENSES.md for dependency licenses.