Auto Mode

Normally you pick an agent tier — Lite, Core, Plus, or Pro — before you send a message, and every message in that chat runs on it until you change it. Auto mode hands that choice to the agent: for each message you send, it looks at what you’re asking and runs on the tier that fits, picking the least expensive one that can do the job well. So a quick “change this label” might run on Lite, while “add a booking flow with availability and payments” in the same conversation runs on Pro, and you never touch the selector. Auto mode works with both the Developer Agent inside a project and Starcat across your organization. It’s experimental and off by default; you turn it on yourself.

​

Turning it on

Open the model selector in the chat input (the button showing your current tier and mode) and switch on Auto. While it’s on:

• The tier list is disabled and the selector reads Auto. You’re no longer choosing a fixed tier.
• Your choice is remembered for that project (or for Starcat), so Auto stays on across those chats until you turn it off.

To go back to choosing the tier yourself, switch Auto off and pick a tier as usual.

​

How Auto picks a model

For every message, Auto reads the conversation so far, not just your latest line, and picks the cheapest tier that can handle what you’re asking. A short request inside a complicated app can still get a strong tier when the task warrants it.

• It steps up the moment you need it. When your next message clearly calls for more reasoning than the last one used, Auto moves to a stronger tier right away.
• It avoids extra model switches. Just after a reply, staying on the same model is faster and cheaper because the conversation is still cached on it. For a short window Auto prefers to keep that model rather than drop to a cheaper one for a marginal saving. It always steps up when the task needs it.
• It respects your credits. If your balance is low for the tier it wanted, Auto picks a cheaper tier you can afford instead of stalling.

If you also have auto-escalate turned on, the two work together: Auto chooses the tier each message starts on, and auto-escalate can still bump it to a stronger model mid-task if a cheaper tier gets stuck.

​

Cost and transparency

Each message is billed as if you’d picked its tier yourself. Auto changes which tier runs, not how tiers are priced, and the routing step that makes the decision adds only a small amount per message. Every reply is tagged with the tier that produced it, so you can scroll back and see which model handled each message.

​

The Multi-lingual model

Next to Auto you’ll find a separate Multi-lingual option, which appears when Core is selected. It swaps in a model that’s stronger at producing non-English text — Thai especially — and at reading images and other visual input. Reach for it when the main job is writing in another language: drafting Thai emails or marketing copy, translating documents, or filling in non-English UI text. For debugging or general engineering work, leave it off and use a normal tier even when your app or content is in another language.

​

When to use Auto

Auto is most useful in longer sessions where the difficulty varies from message to message, or when you’re not sure which tier a task needs. Pick a fixed tier instead when you want a guaranteed level for the whole chat (for example, keeping a long build on Pro from start to finish) or when you’re deliberately staying on Lite to keep a session cheap.

​

Related