Models & loadout

Your loadout is the short list of models you move between. Each of the first nine picks has its own shortcut, and each model has one card. Reordering favorites changes their shortcuts and keeps the default you chose.

A new chat starts on the model you last picked, in the model picker or with ⌃⌘1…9. Until you pick one, it starts on your default, then the first favorite. When the model you last picked is out of usage or signed out, the new chat starts on the next favorite that can run. A card above the composer, Starting on …, says why, for example “Opus 5 is out of usage until 14:00, so this chat starts on GPT-5.6 Sol.”, and Use Opus 5 puts the chat back on it. Your pick is not changed, so once the limit resets new chats start on it again. A model that is only close to its limit still starts.

Until you change it, your loadout holds up to five of the newest models from Claude Code and ChatGPT, whichever you have installed, starting with your default provider's model.

The loadout works across providers: a Claude Code model, a ChatGPT model and an OpenRouter model can sit side by side, and switching between them is one keystroke.

The model picker

Click the model chip at the right of the composer, type /model, or press ⌘/ (Chat ▸ Change Model…; ⌘⇧M works too).

• Each row is one pick: the provider’s icon and the model. The check marks the model the chat runs; the other rows show their shortcuts. A favorite whose plan is nearly spent says so before its shortcut: Weekly 92% when it is close, Out · back 14:00 or Out of usage when it is spent, and Sign in when its login is signed out. You can still pick it.
• Click a row to switch the chat to it. The picker restores that model’s last options for the selected account. For a model you have not configured, a compatible effort carries over; otherwise it uses the model’s initial default. Choosing the current model keeps its options. The picker closes.
• Type in Search models to search your favorites and then every provider’s models, with favorites listed first. The star adds or removes a favorite without switching the chat; adding a model that is already a favorite keeps its card. Large lists scroll inside the picker. Remote catalogs come from the selected Mac.
• Run on several models opens its own panel, where you choose the models that run your prompt with the chat’s own and whether they run Side by side, each in its own worktree, or As a team in the chat’s folder. Once two or more are set, the row says so, for example 3 models · Side by side. Turn off Several models to go back to one model.
• Edit models opens Settings ▸ Default models, where Set default selects the model for new chats without changing the list order.
• Use ↑ and ↓ to move, → to open a submenu, ↩ to choose and Esc to close.

A row that can’t be used right now is dimmed, and its tooltip says why: “Stop the current turn to change this.” while a turn is running and the pick is on another provider, “The claude CLI isn’t installed.”, “Connect OpenRouter in Settings first.”, or “Wait for this conversation to finish loading.”.

Choosing a pick on another provider moves the chat to that provider, which takes over the conversation from your next message. Handoffs & side chats explains what carries over.

Under the picks are the chat’s own settings. They change the chat and remember the model’s options for that account. Each appears only where it applies:

• Auto, once it is turned on in Settings, which lets Codz choose among your picks. See Auto.
• Effort, with a submenu of the model’s levels, for a model that has levels.
• Fast, the provider’s fast tier, for a model that has one.
• Account, once the provider has a second login. See Providers & logins.

Effort and fast mode

Effort is how hard the model thinks. Each model lists the levels it takes, such as Low, Medium, High, Extra high and Max, with Auto first. Auto uses the effort set for that provider in Settings ▸ Models & providers (Effort for Claude, Reasoning effort for ChatGPT), or the provider’s own default when none is set there. A model that takes no effort level has no Effort row.

Fast turns on the provider’s fast tier, and is offered only where the model has one. Elsewhere the picker has no Fast row.

⌥⌘/ steps the chat through its model’s levels, starting at Auto, and ⌘⇧E turns fast mode on and off. Neither does anything on a model without levels or a fast tier. The picker shows each shortcut on its row. Changing effort or fast mode keeps the check on the model’s row. Switching away and back restores those choices, including an explicit Auto selection. A model without effort or a fast tier hides those controls without erasing the other model’s choices. Fast is only restored for the model where you enabled it; it is not carried to a different model.

The model chip shows the model and its effort, with a bolt while fast mode is on. It names the provider in its tooltip, with where the model came from. When nothing has named a model yet, the chip says, for example, Claude default: the chat runs whatever the provider picks.

Keyboard shortcuts

All four also work on a new chat before you send it, where they set up the chat you are about to start. Chat ▸ Switch Model lists the picks by name, with their keys.

Edit the loadout

Open Settings ▸ Default models. Selected models holds your picks as cards, with your chosen default marked Default. All models has a column for each provider: Claude Code, ChatGPT, Cursor, OpenCode, Ollama and OpenRouter, and DeepSeek once it is set up.

• Click a model in a column to add it at the end. Use its remove control to take it out. Drag a model onto a card to replace that pick.
• Add a model, after the last card, lists each ready provider’s models that aren’t in a card yet.
• Drag a card to reorder, or drag it off the row to remove it. The ✕ on a card also removes it. Removing the default selects the first remaining favorite; the last card can’t be removed.
• Click a card to change its Model, initial Effort or Fast. A new chat uses them on that model until you change effort or fast mode in a chat; after that it uses your last choice for the model. A loadout saved with two cards for one model keeps the first.
• Make default changes where new chats start until you pick a model, without moving the card.
• Reset goes back to the models Codz picks for you.

A provider’s column lists its models, models you have used lately, and the default model from its settings. Search models, beside All models, filters every column at once. An empty OpenCode column fills in once OpenCode has run a model. A provider that isn’t ready offers a button such as Set up Claude Code, which opens Settings ▸ Models & providers.

Until you first edit it, the loadout is built from your settings: the default model of your default provider, then Claude Code’s and ChatGPT’s when their CLIs are installed. There is no limit on its length; only the first nine picks get shortcuts. Settings ▸ Agent & composer ▸ Default model names the first pick, with Edit loadout.

Auto

Auto lets Codz choose which of your picks answers each message. A model called Jev, reached through your OpenRouter account, reads the message and picks one. Each choice costs a fraction of a cent, billed to that account.

To use it:

1. Connect OpenRouter in Settings ▸ Models & providers (see Providers & logins).
2. Turn on Settings ▸ Agent & composer ▸ Pick the model automatically.

New chats then start on Auto, and an Auto switch appears in the model picker to turn it on or off for each chat. Until OpenRouter is connected the switch is dimmed, with Needs OpenRouter beside it.

• Auto excludes Ollama and chooses among your other loadout picks. When only one of them can run, it asks nothing.
• To decide, Codz sends OpenRouter your message, with any code you added to it, a line or two from each of the chat’s last few turns, the chat’s mode and current model, the names of attached files, and whether an image is attached.
• When Auto changes the model, a line above the reply names the pick and how confident Auto was: “Auto picked” and the model. It needs to be more sure to move a chat to another provider in the middle of a task.
• When Auto keeps the model, nothing is added to the transcript. The model chip’s tooltip says what happened, such as “Auto kept this model” or “Auto could not reach Jev, so this chat stayed on its model”.
• If Jev can’t be reached or doesn’t answer in time, the message runs on the chat’s current model.
• While Auto decides, the send button dims and pulses, and its tooltip says “Choosing a model”. A message sent meanwhile waits its turn.
• Choosing a model yourself turns Auto off for that chat. While Auto is on, no row in the picker is checked, and the chip names the model that ran last, with Auto beside the effort.

When a plan runs out

When Claude Code or ChatGPT refuses a turn because your plan’s usage limit is spent, the turn stops with the provider’s notice at that point in the transcript. The composer says You’re out of Claude usage (or Selected model is out of usage when only that model is spent), with when it resets and, when the provider names a page, a button such as Add Credits, Upgrade or Raise Limit. The banner goes once the limit resets or you send another message.

With Settings ▸ Agent & composer ▸ Continue on another provider on, as it is by default, Codz doesn’t stop there. It moves the chat to the next ready model in your loadout and asks it to pick up where the previous agent stopped. When a usage limit hits explains the order Codz tries, and Plan limits covers the meters behind the limits.

Troubleshooting

A pick is dimmed. Its tooltip gives the reason. Stop the running turn before switching provider, set up the provider (see Providers & logins), or wait for the chat to finish loading.

A ⌃⌘1…9 shortcut does nothing. Your loadout has no pick in that slot, or the pick can’t run right now.

There is no Effort or Fast row. The model has no effort levels, or no fast tier. Pick another model, or leave it as it is.

There is no Auto row, or it is dimmed. Turn on Pick the model automatically and connect OpenRouter, as above.

Settings shows “Unknown model” under OpenRouter. The Default model id isn’t among the models OpenRouter lists as able to use tools, and chats on it will fail until you correct it.

A ChatGPT turn says “ChatGPT needs an update”. The model needs a newer Codex CLI. Click Update ChatGPT…, then send the message again.

Ollama models. Download or import a model in Models & providers, or select one from the live cloud catalog. Every Ollama pick names its Local, Ollama sign-in, or API key destination. Model metadata controls available thinking levels, Agent mode, and image support.