The same agents you run in the terminal, in a Mac app for non-engineers. Runs on your existing Claude or ChatGPT subscription. Free.
Nova is a workspace for working with AI agents on real knowledge work: PRDs, notes, research, decisions, Slack threads. It drives the same Claude (or ChatGPT) you already run in the terminal and gives it a window you can actually read and edit.
Built for the people who don't write code all day: product managers, founders, writers, researchers, ops.
- Projects and chats in a left panel, chats grouped by folder or status.
- Your files and a real terminal a glance away on the right.
- Markdown, fully rendered and editable in place. Read and edit like a human does, not a computer. No asterisks, hashtags, or brackets. Autosaves as you type.
- Live artifacts. HTML and JSX render live in a sandboxed preview beside the chat.
- A readable chat. You and the agent in clear lanes. Tool activity folds into one tidy card. No more walls of terminal output.
- Your project's local
/skillsin the chat. Type/and the skills in your project's.claudefolder autocomplete. No trip to the terminal. - Claude or ChatGPT. Pick your model right in the composer, per chat, on your own subscription.
- An in-app web browser. Browse and act on live web URLs beside your work — no more tabbing out.
Full walkthrough, screenshots, and the capability-by-capability comparison: supernova.md.
Grab the latest Nova.app / DMG from the Releases page, open it, and drag Nova to your Applications folder. The app keeps itself up to date automatically.
- We never read your conversations. We never see, store, or train on them, and don't want to.
- Requests go straight to your model. Your Mac talks directly to Claude or ChatGPT. Nothing routes through our servers.
App state lives in ~/.nova/; your documents stay as real files in each project folder.
How Nova stacks up against the tools it borrows from, as I used each in mid-2026:
| Capability | Codex app | Cowork app | Claude Code | Cursor | Obsidian | Nova |
|---|---|---|---|---|---|---|
| Markdown rendered & editable in place | – | – | – | – | ✓ | ✓ |
| Readable agent chat, not a wall of text | ✓ | ✓ | – | ✓ | – | ✓ |
| Works in your local files | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ |
| Live artifact preview beside the agent chat | ✓ | ✓ | – | – | – | ✓ |
Your project's local /skills in the agent chat |
– | – | – | – | – | ✓ |
| Runs on your Claude subscription | – | ✓ | ✓ | – | – | ✓ |
| Runs on your ChatGPT subscription | ✓ | – | – | – | – | ✓ |
| In-app web browser | ✓ | – | – | ✓ | – | ✓ |
- macOS. Windows support is on the roadmap.
- The Claude (and optionally Codex) CLI installed and signed in. First-run setup walks you through it.
- Free to use. Bring your own Claude or ChatGPT subscription (or an API key). Driving the agent programmatically draws from the metered programmatic credit, not interactive subscription limits.
Pre-1.0. Silent auto-update is rolling out. The application source is private; this repo is the public distribution channel (downloads + the auto-update feed).