If your team already pays for ChatGPT, your Dock agents can now run on that plan. Since September 11, any agent you've put on GPT-6 Astra or GPT-5.6 can do their work on your own ChatGPT subscription and leave your Dock allowance alone. You flip one setting in the Mac app, and the agents keep their names, jobs and memory.
That was the one release this week, and it's logged in the changelog entry. It landed two days after we wrote up what Dock is for, and a week after Fable 5.1 and GPT-6 Astra arrived in the model menu. Around it sat a busy week of chat fixes you'll notice the first time you watch an agent work.
Your ChatGPT plan, working as a team
Until this week, a Dock agent on an OpenAI model spent your Dock allowance no matter what you were already paying OpenAI. Now you choose who pays.
Open Settings in the Dock Mac app, go to Billing, and expand Bring your own plan. There are two rows, one for Claude models and one for OpenAI models. On the OpenAI row, pick "My Codex plan" and connect. The setting says Codex, and the account it signs in with is your ChatGPT one. That sign-in happens on your Mac and never leaves it. From then on, agents on GPT-6 Astra or GPT-5.6 run on your ChatGPT subscription and use none of your Dock allowance. The row always shows which source is paying right now.
Leave the row on "Dock usage" and nothing changes. Dock never moves you onto your own plan unless you pick it.
Here's what that looks like at a six-person marketing agency where two partners already have ChatGPT plans. Their writer agent stays on Claude, where the house style and client notes have built up over months. Their research agent moves to GPT-6 Astra. With the OpenAI row switched to their Codex plan, the research agent's Monday competitor sweep runs on a subscription the agency was paying for anyway, and the writer keeps working next to them as before.
Switching an agent's model halfway through a conversation also got safer. Move an agent from a Claude model to GPT-6 Astra mid-thread and the conversation comes with them: what you said and what they answered carries across. The agent is told it ran on a different model, so they ask about earlier work rather than pretend to remember steps they can't see.
If you're setting up Dock for the first time in the Mac app, setup now offers your ChatGPT plan as one of the ways to start, and the first agent it creates is put on GPT-6 Astra so the plan can actually pay for them. As of September 13 the option is open to every company on Dock. The plan cards in Settings, under Billing, now say it in one line too: Plug in your Claude or OpenAI plan.
Two limits. Your ChatGPT plan pays for agents running on your Mac; agents running in the cloud always use Dock usage. And it covers GPT-6 Astra and the GPT-5.6 family, while the other OpenAI models in the menu run on Dock usage.
Also shipped this week
- Under each agent reply there's now a one-line bar that reads "N actions". Open it and every step is listed with the specifics: the file they read, the search they ran, the page they fetched, with any secrets in a command masked. When a step goes wrong, the row says so in words, such as "This didn't work."
- Messages you send while an agent is mid-job now wait above the message box, marked Queued, and go in when the current job finishes. Send now stops the current job and sends yours straight away.
- The three-dots menu on an agent in the left rail has Pin to top, which keeps that agent at the top of their group.
- New agents get a face from The Crew, a new pack of pixel-art office workers, and you can change it when you hire. Existing agents keep the avatar they had.
- Pausing or renaming a recurring schedule could quietly turn it into a one-time job. A weekly report now stays weekly.
- Buying more usage from the model menu works. Picking an amount there used to close the window and charge nothing; it now opens payment the same way Settings does.
Every release is logged in the Dock changelog, with the full notes for this week in Run Dock agents with your Codex plan.