VibeStarter AI agents
Five agents and a Terminal tab, as they ship in the app.
An agent is a coding assistant that works in your project: it reads and writes files, tests your game in Roblox Studio, and creates assets. VibeStarter runs each AI vendor’s official tool, not a homemade copy. You do not install anything yourself: the app downloads each tool from its vendor’s official source, and keeps it up to date.
Who does what
Section titled “Who does what”Two accounts are involved, and they do not pay for the same things. Your agents’ reasoning runs on your AI account, with your quota. VibeStarter AI credits only pay for asset generations made by VibeStarter tools. The details are in AI credits, in practice.
The five agents and the Terminal
Section titled “The five agents and the Terminal”| Agent | Account required | Image made by the agent itself | Good to know |
|---|---|---|---|
| Claude Code | Your Claude subscription (Pro or Max) | No | Its images go through VibeStarter tools, so they use your credits or your Gemini key |
| Codex (OpenAI) | Your ChatGPT subscription (Plus or Pro) | Yes, with its own tool | It is also the possible engine of the AI Studio, on your ChatGPT account |
| Grok Build (xAI) | Your Grok subscription | Yes, with a compatible xAI plan | The app says: “Native image tools require a compatible xAI plan.” |
| Antigravity (Google) | A Google account | Yes, with its own tool | The only route announced as free: “Free daily agent usage with a Google account, no credit card required” |
| OpenCode | The provider you choose in OpenCode | No | Its images go through VibeStarter tools, like Claude Code |
| Terminal | None | — | A Linux shell without an agent, in the app’s environment. It is also used to connect Antigravity |
Each agent has an identity color in the header of its panel and in the notifications panel.
The first agent whose connection is confirmed becomes your favorite agent: its icon is pinned to the project bar, and it is the one the app suggests first on an empty tab. Nothing stops you from connecting several and running them together. See Getting around the interface.
To connect one of them, go to Connect your AI account. To choose which one to get, go to Which AI subscription to choose.
What VibeStarter does, and what it does not do
Section titled “What VibeStarter does, and what it does not do”VibeStarter does: install and update the command-line tool of each agent, plug the VibeStarter tools into it, drop Roblox instructions and know-how into your project, and run several agents side by side.
VibeStarter does not:
- Sell AI coding tools. The subscription stays with its vendor, at its price, on your account.
- Choose the model. There is no model picker in the agent panels or in the
presets. The choice is made in the agent’s own tool, with its own
/modelcommand. - Pin a tool version. First launch, repair, and Update all call the current official installer of each vendor.
- Display your quota. It no longer does: your remaining quota is shown on your provider’s website.
The connection sign-in belongs to your provider’s official tool: VibeStarter never uses a subscription token to call a provider on your behalf.
Limits
Section titled “Limits”- One account at a time per agent. You cannot connect two Claude accounts to the same agent in parallel.
- One agent drives a Roblox place at a time. The others wait their turn in Studio and keep working on the rest of their tasks in the meantime.
- The number of agents depends on your computer’s memory, not on an option you buy. See Check that your computer is compatible.
- Agents do not read your game’s statistics. This is not implemented yet.
Frequently asked questions
Section titled “Frequently asked questions”“Is VibeStarter an AI?” No. It is the environment that puts your AI to work, with Roblox tools around it. Without a connected AI account, no agent reasons.
“Do I need all five?” No. One connected agent is enough to work. You can connect several and run them together on different tasks.
“Can the agent be wrong about what VibeStarter can do?” Yes, it happens: an agent sometimes says a feature does not exist when it is shipped. In that case, give it the instruction again explicitly, or clear its conversation context and restart it.