Skip to main content

​
One-line install

curl -fsSL https://augent.app/install.sh | bash
Works on macOS and Linux. Handles all dependencies, installs Augent, and configures MCP. The rest of this page breaks down exactly what runs, provides a manual alternative, and documents how to fully remove Augent.

​
What the installer does

The installer is a single bash script. The source is public and auditable: install.sh. It runs three phases:
1

Phase 1: System dependencies

Detects your OS and package manager, then installs anything not already present. Every dependency is open source and widely adopted:
PackageWhat it ismacOSLinux
Python 3.12Language runtimebrew install python@3.12apt install python3 or equivalent
FFmpegIndustry-standard audio/video toolkitbrew install ffmpegapt install ffmpeg or equivalent
yt-dlpMedia downloaderbrew install yt-dlppip install yt-dlp
aria2Multi-connection download acceleratorbrew install aria2apt install aria2 (macOS and Debian only)
espeak-ngSpeech synthesis phonemizerbrew install espeak-ngapt install espeak-ng (macOS and Debian only)
HomebrewmacOS package managerInstalled if missingN/A
Already-installed packages are skipped. Version requirements: Python 3.10–3.13, no minimum for the rest.On Fedora, the installer also enables RPMFusion to provide FFmpeg.
2

Phase 2: Augent + Python packages

Installs Augent and all Python packages via pip from the GitHub repo. Every package is open source and installed from PyPI:
PackageWhat it isSource
faster-whisperOpenAI Whisper reimplementation, optimized for speedSYSTRAN
PyTorchML framework used by the models belowMeta / PyTorch Foundation
sentence-transformersText embeddings for semantic searchUKP Lab
pyannote-audioSpeaker diarizationCNRS research lab
KokoroText-to-speech (54 voices, 9 languages)hexgrad
DemucsAudio source separationMeta Research
FastAPIWeb framework for the local UI (optional web extra)Tiangolo
Also downloads the audio-downloader shell script to ~/.local/bin/.If the full install fails (common on some platforms due to ML package constraints), it falls back to installing the core package first, then each optional extra individually, reporting what succeeded and what was skipped.
3

Phase 3: Configuration

  1. PATH: if ~/.local/bin or your Python user bin directory isn’t in your PATH, an export line is appended to ~/.zshrc or ~/.bashrc
  2. Package verification: imports each installed module to confirm nothing broke during install
  3. Speaker models: downloads pre-packaged pyannote speaker diarization models (~29MB) from GitHub Releases to ~/.cache/huggingface/hub/. Skipped if already cached
  4. MCP registration: runs claude mcp add augent -s user to register the MCP server with Claude Code. If Claude Code isn’t installed, it prints the command for you to run later. If OpenClaw is detected, it also configures MCP and installs the Augent skill there
When piped from curl, MCP configuration runs automatically. When run interactively (e.g., bash install.sh from a local clone), it prompts first.
After installing, you can optionally create ~/.augent/config.yaml to customize default model size, output directories, and more. See Configuration.

​
What the installer does not do

No root/sudo on macOSHomebrew manages its own directory permissions. The installer never calls sudo on macOS.
No background processesNo daemons, launch agents, systemd services, or cron jobs are created.
No telemetryNo analytics, no tracking, no phone-home. All network activity is package installation from GitHub, PyPI, and Homebrew. On Fedora, RPMFusion is also contacted for FFmpeg.
No system file modificationsOutside of standard package manager operations on Linux, the installer only writes to your home directory (~/.local/bin, shell rc file, ~/.cache/huggingface/, Claude Code config).
On Linux, sudo is used strictly for system package manager commands (apt install, dnf install, etc.): the same commands you would run by hand.

​
Manual install

Pick your platform. Each tab is a complete flow from start to finish.
1. System dependencies
brew install python@3.12 ffmpeg yt-dlp aria2 espeak-ng
2. Augent
pip install "augent[all] @ git+https://github.com/AugentDevs/Augent.git@main" --break-system-packages
--break-system-packages is required on Homebrew Python due to PEP 668. This is safe: Homebrew Python is not your system Python.
3. audio-downloader
mkdir -p ~/.local/bin
curl -fsSL https://raw.githubusercontent.com/AugentDevs/Augent/main/bin/audio-downloader -o ~/.local/bin/audio-downloader
chmod +x ~/.local/bin/audio-downloader
Add to PATH if not already present:
echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc
source ~/.zshrc
4. MCP server
claude mcp add augent -s user -- python3 -m augent.mcp
5. Verify
augent --help
audio-downloader --help
Restart Claude Code and run /mcp to confirm.

​
Uninstall

Remove Augent:
pip uninstall augent -y
rm -f ~/.local/bin/audio-downloader
claude mcp remove augent -s user
rm -rf ~/.augent
Remove speaker models installed by the installer:
rm -rf ~/.cache/huggingface/hub/models--pyannote--speaker-diarization-3.1
rm -rf ~/.cache/huggingface/hub/models--pyannote--segmentation-3.0
If you used OpenClaw:
rm -rf ~/.openclaw/skills/augent
Also remove the augent entry from ~/.openclaw/openclaw.json if present.
Other ML models (Whisper, sentence-transformers, Kokoro) are downloaded on first use and cached under ~/.cache/huggingface/hub/. These are shared with other HuggingFace tools. Only remove them if nothing else on your system uses them.