Accounts and models

Zero keeps accounts for Claude Code, Codex and OpenCode. Each time a task starts, Zero gives one account to the agent. The agents talk to their providers themselves; Zero never does. Your keys and tokens stay in your Keychain.

Accounts in Zero are optional. When Zero has no account for an agent, the agent’s tasks use the sign-in that the agent already has, the one it uses in Terminal. After you add an account for an agent, its tasks use Zero’s accounts.

Open the Accounts page

Open Settings › Accounts. To open Settings, press ⌘ , or click the gear at the foot of the window’s left rail.

The page has a section for each agent: Claude Code, Codex and OpenCode. Each section lists its accounts. A section with no accounts shows None yet. Add model and Add account are beside the agent’s name.

Add an account

Click Add account beside the agent’s name and choose how to sign in:

• If Claude doesn’t say whose account it is, the new row asks Whose account is this? (its email, or a name). Type one and press Return.
• A ChatGPT sign-in goes away when you refuse it, or when you don’t finish it within 15 minutes. To try again, add it again.
• If a Claude sign-in fails, the dialog says why and shows Try again.

Under each account’s name, its row says what it is:

• Its plan, as the agent reports it: a Claude plan, or ChatGPT and its plan.
• Subscription, when Claude doesn’t say which plan.
• API key. For OpenCode, the name of the key’s variable comes after it.

Rename or remove an account

Click ⋯ at the end of the account’s row to open its menu:

• Rename: type a new name and press Return. Esc cancels.
• Switch to a token: shows for a Claude account whose row says its own folder: your MCP sign-ins aren’t shared. It opens Claude’s sign-in again. With the token it gets, the account shares your MCP sign-ins.
• Remove…: the row asks again. Click Remove to confirm, or Cancel. The tasks of a removed account can’t restart with it. To get the account back, sign in again.

See how much usage is left

Click an account’s row to open it in place. It shows a tile for each allowance its plan has:

• Last 5 hours: how much of the five-hour allowance is used.
• This week: how much of the weekly allowance is used.

Each tile shows the share used, a bar and when the allowance resets. The bar turns orange at 90% and red at 100%. An account at a limit shows At its limit on its row. Click the row again to close it.

Under the tiles, Activity has a square for each day, darker the more turns the account’s tasks finished that day. The count starts on the day Zero began to keep it.

While the page is open, Zero asks each agent for its accounts’ usage every two minutes. API keys have no allowance, so they show no tiles.

The foot of the tasks sidebar also shows the accounts your tasks used in the last day, with a 5h and a Week bar each. Point at a bar to see when it resets. Click the card to open Settings › Accounts.

Which account a task uses

You don’t pick a task’s account. Zero picks one when the task starts:

• A worker of the same agent uses the same account as the task that started it.
• For other tasks, Zero leaves out accounts that are at a limit, and accounts that can’t use the task’s model. From the rest, it picks the account with the least weekly allowance left. So Zero uses up one account before it goes to the next, and the other accounts keep their allowance. When two accounts are equal, the one that resets sooner goes first.
• When every account is at a limit, Zero uses the first account in the agent’s list.

When a task changes account

A task keeps its account until one of these happens:

• The account reaches a usage limit. The task continues on another account of the same agent, with its conversation. Zero tells the agent, and the agent continues from where it stopped.
• The account can’t use the task’s model. Zero records that the account can’t use that model, and the task continues on an account that can. Zero doesn’t pick that account for that model again.

When no other account can take over, the task stops. For a usage limit, the task shows Usage limit at the end of the conversation, with the time when the limit resets. After the reset, the task continues by itself. To continue sooner, click Continue with Codex or Continue with Claude Code there. The other agent then takes over the task and its whole conversation.

When an account’s key is missing from your Keychain, the task doesn’t start. It shows “isn’t in the Keychain any more: sign in again”. Remove the account and add it again.

Pick a model and effort for a new task

To open the New task page, click Home, the house at the top of the window’s left rail. On the right of the row under the box, a chip shows the agent’s icon, the model and the effort, such as “Model · High”.

1. Click the chip. It has a tab for each agent: Claude Code, Codex and OpenCode. To switch agent, click a tab or press ← and →.
2. Point at a model, or move to it with ↑ and ↓. The pane beside the list describes the model. For models that have levels, the pane also shows an Effort slider.
3. Click the model, or press Return. Or move the Effort slider. The slider picks the model at that level and keeps the list open. Esc closes the list.

When you pick no model, the task uses the agent’s first model. OpenCode uses the default from its own configuration. The effort levels are different for each model. On the chip, Zero shows Extra High as “XHigh”.

Zero remembers the agent you picked last and uses it for the next new task.

Change the model during a task

In a task, the same chip is in the composer, at the foot of the conversation.

• Between turns, the model and effort you pick apply from your next message.
• While the agent works, your choice waits for the next turn. The chip shows a mark. When you point at it, it shows From the next turn.

The list also shows the models of the other agents. When you pick one, the task moves to that agent with its whole conversation. Claude Code and Codex can take over a task. OpenCode can’t.

Add a model the list doesn’t show

Sometimes a new model is available before an agent lists it. To add it to Zero:

1. Open Settings › Accounts.
2. Click Add model beside the agent’s name.
3. In Add a Claude Code model, or the Codex or OpenCode form, type the model’s id in the form that the agent accepts. For example, claude-mythos-5-2, gpt-5.5-codex, or provider/model for OpenCode.
4. Optionally, type a name for the list, such as “Mythos 5.2”. Without one, the list shows the id.
5. Click Add.

The model shows under the agent’s accounts and in every model list for that agent. All of the agent’s accounts offer it.

When an account turns out not to have the model, its tasks on that model run on your other accounts. The model’s row then says Not on and that account. When none of your accounts can use the model, the row says None of your accounts can use it. To take a model out, click Remove.

Where the lists come from

• Claude Code: Fable, Opus, Sonnet and Haiku, which Claude Code resolves to the latest model of each. Then the models that Claude Code has offered your account, such as the 1M-context versions, and the models you added. Zero reads these from Claude Code’s own files and never calls Anthropic.
• Codex: the models that Codex offers your account, as Codex lists them, plus the models you added.
• OpenCode: its own list, which Zero reads when you open the chip. The rows fill in after Zero reads the list.