How to connect: which agent, which path
Last reviewed: 28 Oct 2026
Every agent connects to Naetive the same way underneath: through Naetive's MCP server, acting as you with your connection key. What differs is how you get there and what the agent can reach on your machine afterwards.
Connect today
Open Connect in the side navigation. The top of the page always shows one next step:
- Which agent do you use? until your first agent connects. Pick Claude Code CLI, Codex, ChatGPT, Claude, web or app, or Cursor or another, and follow the steps shown.
- Approve your computer while an install line waits in a terminal (its code lasts ten minutes).
- A red card if the console or the tunnel stopped on this computer, with the line that starts it again.
- Give your project its first understanding once an agent is connected, with a line to paste.
- Then one quiet line: who is here now, and an Open Work button. If nothing is installed on the computer you are reading on, it offers Set up this computer for Naetive instead (skip it with Skip on this computer).
Under it: Your agents (each kind you have connected, with Add another agent for the rest), Chats and file access on this computer (optional), and Settings (your key, git activity, and removing Naetive from this computer).
The one-minute path: a chat agent
ChatGPT, Claude on the web, Claude Desktop. Add Naetive as a connector by pasting one address from the Connect page, following the numbered steps there. Nothing to install. The agent can read and write everything on Naetive: the understanding, the work, the documents.
What it cannot do on its own is touch files on your computer. For that, share the project and start the tunnel (below).
The full path: Claude Code
Claude Code is a program on your computer, so it can do everything: work on your real files, report its activity to Work, and refresh the understanding folder in your repo. It needs only the install line from the Connect page, run once from any folder; after that, Claude Code on that computer arrives knowing which project each folder is. See Connecting Claude Code.
The middle path: Codex CLI
Codex runs in your terminal too. It connects through a short config entry (Connect shows it), and reaches your files through the tunnel like a chat agent, or by running on the Chats page.
What each path gives you
| Chat agent (ChatGPT, Claude web, Claude Desktop) | Codex CLI | Claude Code | |
|---|---|---|---|
| Reads and writes the understanding and the work | Yes | Yes | Yes |
| Shows up in Who's here on Work | When it makes a call | When it makes a call | Live, with what it did |
| Works on the files on your machine | Through the tunnel | Through the tunnel | Directly |
Refreshes the .naetive folder in your repo |
No | No | Yes, through the installed hooks (macOS, Linux, WSL) |
| Runs in the platform's Chats page | No | Yes | Yes |
The tunnel, in one paragraph
The tunnel is a small program the install line puts on your computer. It belongs to the computer, not to a project: once you share a project it starts by itself in the background (on Windows you start it yourself: start-tunnel.cmd), and from then on agents that live elsewhere (ChatGPT, Claude on the web) can read and change the projects you shared on that computer, and only those, through Naetive. A project is shared by the checkbox Let ChatGPT and Claude on the web reach this folder when you create it, or by naetive share --project <name> (naetive shares lists them); your files stay on your computer, and what an agent asks for travels to it, encrypted. It survives Naetive's own deploys and refuses a call for a project it does not serve. See The tunnel and Sharing your projects and starting the tunnel.
Your connection key
Your account has one connection key, and each computer and each agent you set up by hand gets its own.
- Each computer gets its own key when you approve it during the install line. Claude Code, the console, the tunnel and the
naetivecommand on that computer use it. Running the install line again there replaces it. - ChatGPT, Claude on the web, Codex, Cursor and other agents get their own key from their steps on Connect (Make a key for …). It is shown once, inside the address or setting you paste, so copy it straight away and share it with no one. Making another replaces only that agent's key. (Where Connect offers it, ChatGPT and Claude can sign in to Naetive instead of using a key.)
- The account key is under Settings, Your connection key on Connect, shown with only its last four characters. You rarely need it.
- Naetive holds only a fingerprint of each key, never the key itself.
If a key may have leaked: for an agent's own key, make a new one for that agent, and the old one stops working at once. For a computer's key, run the install line again on that computer. For the account key, click Regenerate, then Yes, regenerate the key: the old key stops working at once, including tunnels and terminals already running on it. Computers and agents with their own key are not affected.
Which should I start with?
Start with the agent you already use most. Most people connect a chat agent in a minute to see the understanding arrive, then run the install line for Claude Code when they want an agent working on the real code.