LuaMind: AI agent for roblox studio

Resources Community Resources
, , , ,
1

LuaMind

It’s the only creative name I could think of. If you have any recommendations, please let me know :>.

LuaMind is an advanced AI agent integrated into Roblox Studio. Unlike a conventional chatbot, it operates contextually and autonomously by executing logical tools for reading, diagnosing, and manipulating the environment (DataModel).

To ensure a safe and efficient workflow, the system divides its capabilities into Safe Read actions and Destructive/Modifying actions that require developer consent.

:shield: Safety and Control Policies (Safety System)

The agent has a strict layer of data protection and environmental control. Tools classified in the table below under the status :warning: Requires Confirmation will pause the AI’s automated process and trigger an interactive alert in the interface for the user to validate the change before it is applied in the game.

Tool Name Action Type Security Status Functional Description
get_file_tree Read Safe Retrieves the game hierarchy tree starting from a specified root (default: game) with configurable depth limit
get_instance_properties Read Safe Reads and displays common properties (Name, ClassName, Position, Size, Color, etc.) of a specific instance
get_instance_children Read Safe Lists all direct child instances of a specified parent object
get_script_source Read Safe Retrieves the full source code from a Script, LocalScript, or ModuleScript
set_script_source Write Destructive (Requires Confirmation) Replaces the entire source code of a script with new content
create_instance Write Destructive (Requires Confirmation) Creates a new Roblox instance (Part, Script, etc.) and parents it to the specified location
delete_instance Write Destructive (Requires Confirmation) Permanently removes an instance and all its children from the game
set_properties Write Destructive (Requires Confirmation) Modifies one or more properties of an instance (position, size, color, etc.)
search_objects Read Safe Searches for instances by partial name match and/or class name filter within a root container
get_selection Read Safe Returns all instances currently selected in the Roblox Studio viewport
run_code Execute Destructive (Requires Confirmation) Executes arbitrary Luau code in the plugin context and returns the result
insert_model Write Destructive (Requires Confirmation) Loads and inserts a model from the Roblox catalog using an Asset ID
get_output_log Read Safe Retrieves recent messages from Studio’s Output window (configurable line limit)
rename_instance Write Destructive (Requires Confirmation) Changes the Name property of an existing instance
move_instance Write Destructive (Requires Confirmation) Reparents an instance to a new parent location

Why LuaMind Special
Plugin doesn’t use a single, fixed AI. You provide your own API key and endpoint. This offers:

  • Multi-vendor support: Switch between Cerebras, Groq, Google, and NVIDIA without touching the code; simply add a new endpoint.
  • Key per model: Each model stores its own key. You change the model, but you don’t rewrite the key each time.
  • Confirmation before breaking something: Destructive tools (create, delete, write script) stop and ask for confirmation/cancellation. Read tools run automatically, without interruption.
  • Real agentic loop: The AI ​​can chain several tools in a single message (read tree → read script → rewrite) without you sending message by message.
  • Automatic retry at the rate limit: If it encounters a 429 error, it waits and retries automatically; it doesn’t crash.

List of common errors + why they occur

  • HTTP 400 (Bad Request)
    This occurs when: the model name is misspelled (incorrect capitalization, incorrect hyphen), the submitted body does not conform to the schema required by the provider (for example, Cerebras rejects empty required:[] in the tools’ JSON schema), or the model is not supported by the plugin’s tools.

  • HTTP 401 (Unauthorized)
    The API key is empty, incorrectly copied (with a space or extra quote), or expired/revoked in the provider’s dashboard.

  • HTTP 429 (Too Many Requests)
    The provider’s rate limit has been reached (X requests per minute/day). The plugin will retry automatically, but if the problem persists, you need to wait longer or change providers.

  • “HttpService is not enabled”
    The permission to allow HTTP Requests must be enabled in Game Settings → Security → Allow HTTP Requests, within Studio.

  • “Content-Type is not allowed”
    Previously, this required manually sending the Content-Type header along with Enum.HttpContentType.ApplicationJson (Roblox already does this automatically; duplicating it throws an error). This has been fixed in the current version.

  • Button remains stuck in cancel mode.
    This happened if you canceled just as AI was about to display the final result. Fixed: now it checks for cancellations at that exact moment as well.

  • “Select provider before model”
    You were adding a model without first saving/selecting the provider with its endpoint.

How to add a model and provider (Cerebras example)

  1. Open the settings panel (:gear:).
  2. In the provider name field, type: Cerebras AI
  3. In the endpoint field, paste: https://api.cerebras.ai/v1/chat/completions
  4. Click the “Add Provider” button.
  5. In the “New Model” field, type the exact name: gpt-oss-120b
  6. Click “Add Model” — it will be automatically linked to Cerebras.
  7. Paste your Cerebras API key in the Keybox and save.
  8. Open the models dropdown at the top and select the one you just created.
  9. Done, send a message.

The 4 Recommended Providers (Order Best to Worst,in my opinion)

  1. Cerebras AI — Best for Heavy Daily Use

URL: https://cloud.cerebras.ai/
Endpoint API: https://api.cerebras.ai/v1/chat/completions
Get a key: Create an account, go to the API Keys section, generate a new one, and copy it.
Features: 30 requests/minute, 14,400 requests/day, 1 million tokens/day, no credit card required. Highest volume of all.

  1. Groq Console — Fastest

URL: https://console.groq.com/
Endpoint API: https://api.groq.com/openai/v1/chat/completions
Get a key: Register, Dashboard → API Keys → Create API Key.
Features: 30 requests/minute, 1,000 requests/day, built-in hardware LPU processes 300-1,000 tokens/second, no credit card required.

  1. Google AI Studio — best long context

URL: https://aistudio.google.com/
Endpoint API: https://generativelanguage.googleapis.com/v1beta/openai/chat/completions
Get a key: Sign in with your Google account, click “Get API Key,” and create a project if prompted.
Features: 15 requests/minute, 1,500 requests/day, context up to 1 million tokens, no credit card required. Gemini 2.5 Flash recommended.

  1. NVIDIA NIM — Various Frontier Models

URL: https://build.nvidia.com/
Endpoint API: https://integrate.api.nvidia.com/v1/chat/completions
Get a key: Create an account, search for a model, and click “Get API Key”.
Features: 40 requests/minute, no fixed daily limit but with limited credits that run out, no credit card required. Good variety of models (Llama, Nemotron, etc.).

Plugin: https://create.roblox.com/store/asset/93694815642868/LuaMind

Hablo español nativamente,ingles tovia no se hablarlo muy bien.