Connect Claude, Cursor or Codex

Open the agent → AI Model (under General). The screen lists the models available to the agent — leave Default (recommended) or pin one — and below it you can connect one of these subscriptions at a time:

Harness	You need	Runs as
Claude	A Claude Pro or Max subscription	Claude Code in the agent’s container
Cursor	A Cursor account	Cursor CLI in the agent’s container
Codex	A ChatGPT account	OpenAI Codex in the agent’s container

While a harness is connected, responses run on your plan and TeleClaw credits are not spent on them. Connecting a harness requires TeleClaw Pro.

Connect Claude

1. Tap Authorize with Claude Code — claude.ai opens.
2. Approve the request. Anthropic shows an authorization code.
3. Copy the code back into the Authorization code field and confirm.

Connect Cursor

Tap Authorize with Cursor, finish signing in on the page that opens. The screen updates by itself once Cursor confirms — you don’t need to paste anything.

Connect Codex

1. Tap Authorize with ChatGPT. TeleClaw shows a device code.
2. Copy it, sign in to ChatGPT and enter the code.
3. The screen updates automatically once ChatGPT confirms.

Device-code sign-in must be allowed

Codex uses device-code sign-in. It has to be enabled in your ChatGPT security settings — and on an enterprise plan, your workspace admin has to enable it first, or the authorization will keep timing out.

Model choice, switching and disconnecting

Under Model you can leave Default (recommended) or pin a specific model the harness offers.

Only one harness can be active. Disconnect the current one before connecting another; disconnecting puts the agent back on TeleClaw’s models and credits. If an authorization times out, start over — the codes are short-lived.