Short answer: the Zetta Agent is a game-development assistant that works inside the Unity Editor. It reads your project before it touches anything, makes changes with Unity's own tools — scripts, GameObjects, components, prefabs, materials — and checks the result by compiling, reading the Console, entering Play Mode and looking at the Scene view. It can also make the 3D models and images your game needs. It comes with the Zetta 3D plugin for Unity, under Window › Zetta 3D › Agent.
How it works
- Understand. Every message carries a picture of your project: Unity version, target platform, render pipeline, color space, input system, scripting backend, packages, open scenes, the hierarchy, your selection and any compile errors. It reads further wherever the task needs it.
- Plan. It works out what to change, following the conventions already in your project rather than its own.
- Build. It makes the changes in the Editor, one visible step at a time in the chat.
- Check. It waits for scripts to compile and fixes its own errors, reads the Console, enters Play Mode, and looks at the Scene or Game view — then tells you what it did and how to try it.
What it can do
| Area | What it does |
|---|---|
| Scripts | Writes new C# and edits existing scripts, then fixes its own compile errors |
| Scenes and objects | Creates, opens and saves scenes; creates, moves, renames, reparents and deletes GameObjects |
| Components | Adds and removes components and sets any serialized property, including references to assets |
| Prefabs and materials | Saves objects as prefabs and creates materials for your render pipeline |
| Testing | Enters and exits Play Mode, reads the Console, looks at the Game view |
| 3D models and images | Generates models, runs retopology, Smart UV, texturing, rigging and segmentation, draws images, then imports and places them |
| Pipelines | Builds a pipeline from a description, runs it and imports what it makes |
How to brief it
- Say what you want, not how. "Make the coins spin when you collect them, and add a score counter in the corner" gives it the outcome; it reads the project to find the how.
- Point at things. Type @ to tag a scene object or a project asset, drag items from the Project window or the Hierarchy onto the chat, or select them and press ⇧⌘M (Ctrl+Shift+M on Windows). A tag is sent as the item's address, and the agent opens it itself.
- Show it. Attach pictures or PDFs, or attach a capture of the Scene or Game view.
- Explore first. Ask mode reads your project and answers questions — explains code, finds where something happens — and changes nothing.
When it needs something from you to carry on, it shows a card in the chat: a slot to drop a prefab or object into, options to choose from, or a text box. It continues the same task when you answer.
Staying in charge
- Everything can be undone. Every change goes on the Undo stack, named after what it did.
- It asks first, by default. It waits for your OK before deleting anything, making or processing a 3D model, or running a pipeline. Full autonomy stops it asking, and it notes every step it approved for itself.
- Stop means stop. Stop, or Esc, ends what it is doing straight away.
Memory
It keeps notes and a brief for each project — what the game is, its conventions, decisions you have made — plus your preferences across projects, and uses them in every conversation. Open Memory to see, edit or delete any of it. Commit ProjectSettings/Zetta3DSettings.json and the whole team shares one project memory.
What it sends, and what it costs
To do a task, the agent sends what the task needs: a summary of your project's setup, the parts of the scene and the scripts it reads, Console messages and any view it captures. It does not upload the project wholesale. The privacy policy covers how that is handled.
Its work uses your AI token allowance, and any model it makes or processes uses credits, as it would on the web. On Auto, each message goes to a model and a depth of thinking sized to the task, and when a job turns out longer than it looked, or keeps hitting errors, Auto steps up mid-task. You can choose Fast, Capable or Intelligent, and a thinking level from none to Max, yourself. While it works, the chat shows which model and thinking level it is on.
Running it on your own API key
On any plan, the agent can run on your own API key from Gemini, OpenAI, DeepSeek or Claude instead of your Zetta account.
1. Open Unity › Settings › Zetta 3D (Windows: Edit › Preferences › Zetta 3D). 2. Under Zetta Agent — account, set Run the agent on to My own API keys. 3. Paste a key next to a provider and click Save. It is checked straight away; Get a key opens the provider's page.
Under the message box, the first menu then lists Zetta Account and every provider you saved a key for. Pick a provider, any of its models and a thinking level — only the levels that model has are offered, and Auto still decides the thinking for each message. The model you pick is used for every message, and Zetta Account is one click away in the same menu.
- Billing. What the agent uses on your key is billed by that provider, not in Zetta credits. Making 3D models and pictures still uses credits.
- Where the key lives. In the macOS Keychain, Windows Credential Manager or the Linux keyring — not in Unity's preferences or your project, so it never reaches version control. It travels with each message, through Zetta 3D to the provider, and is never saved on our servers.
- Pictures and PDFs. For a model that cannot read pictures, Scene and Game turn off, and the agent tells you before sending a file it cannot open.
- When a key stops working — wrong, out of credit, or the model is no longer offered — the chat says so, with Open settings to fix it.