Quickstart

The personal path requires Node.js 20.10 or newer. After installation, moxxy is the only command you need to remember.

1. Install

npm install -g @moxxy/cli

2. Start in a project

Run moxxy from the project you want it to understand:

cd your-project
moxxy

On first start, choose one of the supported connection paths:

• Anthropic or OpenAI API key;
• ChatGPT OAuth, without an API key;
• an existing Claude Code subscription session.

Moxxy stores API keys in its encrypted local vault, selects recommended defaults, and opens the TUI. No config file or separate onboarding command is required.

3. Complete a useful task

Try a read-only first task:

Explain how authentication works in this repository and point to the files that matter.

Then try a small change:

Add a focused test for the invalid-token path and run it.

Built-in reads, searches, and file listing stay quiet while their resolved paths remain inside the workspace. For a write or command, moxxy shows the target and impact first: press Enter to allow once, A to allow that exact consequence for the current run, D for raw details, or Esc to deny.

For one answer in the current shell:

moxxy -p "summarise the README in three bullets"

Continue an earlier run with moxxy resume. Use /runs to switch inside the TUI and /extensions when you deliberately want to add capability.

Diagnose a connection

moxxy doctor --check-keys
moxxy login status

For a Claude Code subscription, sign in with claude auth login before the first start. For ChatGPT, first start opens OAuth. Headless services should authenticate once as the same OS user before they are started.

Add capabilities later

Extensions, channels, automation, and background services are optional:

moxxy extensions list
moxxy onboard --advanced
moxxy help advanced

The desktop follows the same disclosure: Runs, Extensions, and Settings are the primary navigation; collaboration, automation, apps, channels, voice, and mobile live under More.

Next

• Developer alpha — supported paths, limitations, and feedback.
• Permissions — what gates a tool call.
• Authoring a skill — the lightest extension.
• Governed workstations — organization policy and receipts.