Skip to main content
Every target except Notis runs on a coding-agent account you sign in yourself. The two cloud targets are connected inside Notis; the two local ones reuse the CLI login that already exists on your machine. All four are shown in one place: open your account menu and choose Agents routing. Each target has its own status card and, when it is not ready, says what it needs.
Notis never asks for an API key and never accepts one. Every coding-agent target runs on your own Codex or Claude subscription, so usage lands on the account you already pay for. The one credential you ever paste — the Claude setup token below — is a subscription login minted by Claude Code itself, not an API key.

​
Codex Cloud

Codex Cloud signs in with a device code, the same way you would sign a TV app in to a streaming service. Nothing is typed into Notis.
1

Open Agents routing

Account menu → Agents routing → the Codex Cloud card.
2

Start the connection

Press the button on the card. It reads Connect Codex Cloud the first time, Continue authorization while an authorization is still waiting on you, and Reconnect after an attempt expired or failed — all three start or resume the same flow. Notis prepares your cloud computer and installs the Codex CLI there, which takes a few seconds the first time.
3

Approve on OpenAI

The card shows a verification link and a short code, and Notis opens that link for you. Enter the code and approve with the ChatGPT account whose Codex subscription you want to use. The copy button next to the code puts it on your clipboard.
4

Come back

The card flips to Connected. The authorization code is single-use and expires after about 20 minutes — if you miss the window, press Reconnect and start again.
5

Switch accounts later

Press Disconnect on the connected Codex Cloud card. This signs Codex out without deleting your cloud files. Press Connect again and approve with the other ChatGPT account.
Your Codex credential is written by the Codex CLI itself, inside your own cloud computer. Notis only ever handles the verification link and the short code.

​
Claude Cloud

Claude Code has no device-code flow, so the credential comes from a token you mint yourself.
1

Mint a token on your own machine

In a terminal, run:
Claude Code opens your browser, you log in with your normal Claude subscription, and the command prints a token.
2

Paste it into Notis

Account menu → Agents routing → the Claude Cloud card → paste only the token and press Connect Claude Cloud. If your terminal wrapped it onto two lines, paste both lines — Notis joins them automatically. The field is masked, and Notis never shows the token again.
3

Notis verifies it

Notis installs the Claude Code CLI in your cloud computer, stores the token there, and runs a check. If the token is good, the card flips to Connected and the button becomes Disconnect. If Claude refuses the token, the card says so and nothing is stored.
What this means in practice:
  • Notis never sees your Claude login. You log in to Anthropic in your own browser. Notis receives only the token that command printed.
  • The token lives in your cloud computer, nowhere else. It is written to a private file that only your sandbox can read. It is not stored in the Notis database, not shown again after you save it, and not written to any log.
  • It lasts one year. Anthropic issues setup-token credentials with a one-year lifetime, and there is no renewal notice — the first sign that it expired is a Claude Cloud run failing. Re-run claude setup-token, paste the new value, and you are back. If you want a reminder, ask Notis to set one for eleven months out.
  • You can disconnect at any time. Removing the connection deletes the token from your cloud computer.
Anthropic documents claude setup-token as the supported way to authenticate “CI pipelines, scripts, or other environments where interactive browser login isn’t available”. That is exactly what your cloud computer is.

​
Codex Local and Claude Local

The local targets have nothing to connect inside Notis — they reuse the CLI you already use in your own terminal. You need three things:
  1. The CLI installed and logged in. Run codex login or claude once in your terminal and sign in with your subscription. Confirm it works there before expecting Notis to use it.
  2. The Notis desktop app running and signed in, with Your Computer enabled. That toggle is the trust boundary — see Your Computer for the full setup.
  3. The desktop app awake. A sleeping laptop makes a Local target unavailable. Pin Auto if you want Notis to keep answering elsewhere; pin the Local target if you want the request to fail rather than move.
Notis never reads, copies, or transmits your local CLI credential. It asks the CLI to run a task and reads back the result.

​
Reading the status cards

Each card carries a status badge, and often a line of detail under it explaining the badge. Any other badge is passed straight through from the runtime check; the detail line under it says what happened. A target that is not ready still shows up in every routing picker, dimmed and marked NOT CONNECTED. Clicking it opens Agents routing instead of selecting it — or, if you are already there, points you at the cards above — so you cannot accidentally aim a channel at something that cannot run.