Getting Started

1. Download and install

Download the latest build from the GitHub Releases page.

These builds are unsigned. macOS and Windows both object on first launch. Neither download is broken; the steps below get you past it.

macOS: “edytlab is damaged and can’t be opened”

The app is not damaged. macOS quarantines downloads and refuses to launch one Apple hasn’t notarized. The dialog offers only Move to Trash, and right-click → Open will not help. Remove the flag after dragging the app to Applications:

xattr -dr com.apple.quarantine /Applications/edytlab.app

Then open it normally. Once per install, until notarization is set up. If you see a milder “unidentified developer” message instead, System Settings → Privacy & Security → Open Anyway is enough.

Windows: SmartScreen

Click “More info” → “Run anyway” on first launch. The warning stays until the installers are signed.

2. Get an API key

You bring your own API key; it is stored in your OS keychain. With a ChatGPT Plus or Pro plan you can skip the key (a preview): see the last row and “Using your ChatGPT plan” in the user guide.

Provider	Where to get a key	Free tier?
Anthropic (recommended)	platform.claude.com/settings/keys	No — pay-as-you-go
Google Gemini	aistudio.google.com/apikey	Yes — free tier available
Groq	console.groq.com/keys	Yes — free tier available
OpenRouter	openrouter.ai/keys	Yes — free models available
OpenAI	platform.openai.com/api-keys	No — pay-as-you-go
Ollama (local)	No key. Install Ollama, pull a model, and edytlab talks to it on your machine.	Free — runs on your hardware
ChatGPT plan (preview)	No key. In Settings, choose ChatGPT plan and Continue with ChatGPT; edytlab opens your browser to sign in. Available for ChatGPT Plus and Pro. May not be available on workspace plans; that depends on OpenAI and your admin.	No separate bill — it uses your plan's included usage

Cost estimate: a typical 30-minute editing session uses roughly $0.05–0.50 in LLM tokens, depending on the model and the session. Haiku-class models cost about 20× less per token than Sonnet-class models. With the ChatGPT plan there is no per-token bill: usage counts toward your plan, and on Plus the five-hour limit is shared with every other app and tool that uses the plan (not on Pro).

3. Enter your key

1. Launch edytlab.
2. Click the gear icon (⚙) in the top-right corner.
3. Pick your provider.
4. Paste your API key (Ollama needs none). For the ChatGPT plan, choose Continue with ChatGPT and finish signing in in your browser.
5. Optionally press Test: a minimal request checks the key and that the model can call tools.
6. Press Save & Continue.

Keys go in your OS keychain (macOS Keychain, Windows Credential Manager, the Secret Service on Linux), never to disk or to any edytlab server. They share one keychain entry, with the ChatGPT sign-in, so macOS asks for your login password at most once per launch. Your provider and model choices are not secret and are saved in a settings file in the app's config folder. On a Linux desktop with no Secret Service running keys are kept in the kernel keyring until you restart, and the app tells you so, at launch and again at Save if the service went away after the app started.

4. Load your first audio file

• Drag and drop a WAV, MP3, FLAC, Ogg Vorbis, Opus, AIFF or (on macOS and Windows) M4A file onto the timeline. Several files at once make a multi-track session.
• Open button: click “Open Audio” in the empty state or the file menu.
• Chat: type load /path/to/file.wav.

5. Your first edit

Describe what you want in the chat panel. Some prompts to try:

• Remove the silence at the beginning
• Normalize to -14 LUFS
• Take the background hiss out
• Find the silences longer than a second
• Export to /Users/me/Desktop/output.wav

The agent streams its response and shows a badge for each tool call. Every change is a new node in the session graph; nothing is destructive.

6. Undo and redo

Every edit is a node in a directed acyclic graph (DAG), so you can undo and redo freely:

• Ctrl+Z (Cmd+Z on Mac) — undo
• Ctrl+Y (Cmd+Y on Mac) — redo

Unlike a linear undo stack, the DAG keeps all branches: fork a session, try a different edit, and switch between branches in the Graph view. Undo retraces the steps you took, even when an edit returned to an earlier state (mute, then unmute); the Graph view marks that path, and it is saved with the project across restarts.

Keyboard shortcuts

Press ? for the full shortcuts overlay.

Next steps

• User Guide — every feature
• Audio Tools Reference — all 113 tools the agent can call
• Developer Guide — build, test, contribute