Skip to main content
Tomdacat Code v3.3 native preview showing the graphite interface, expanded composer modes, and five-level effort slider Tomdacat Code brings a Claude-Code-style AI coding agent directly into Roblox Studio. It reads and edits your place through a real tool-calling loop, asks before it does anything destructive, and shows you exactly what it’s about to change before it changes it. It lives in two dock widgets inside Studio: a main panel — chat, history, subagents, and settings behind a persistent icon rail — and a side-by-side diff viewer for reviewing script edits before you approve them.
This documentation covers Tomdacat Code v3.3. See the Changelog for the redesigned interface, five-level reasoning slider, higher token limits, and generation-free key verification.

​
Why it’s different from a generic chatbot

It actually edits your place

Tools operate on the live Roblox DataModel — reading the hierarchy, scripts, and properties, and writing changes through Studio’s own undo history.

You stay in control

Every mutating tool call can be set to Ask, Always Allow, or Always Deny per tool. execute_luau always asks, no exceptions.

Streams everything, live

Assistant replies and tool-call arguments fill in token by token, and the Stop button actually stops mid-stream now — no more waiting it out.

Extensible by design

Skills inject domain playbooks into context on demand, and Connectors add optional tool capabilities like icon search — both without touching code.

​
Core building blocks

A familiar chat UI with markdown rendering (including tables), an auto-growing input box, slash commands, @ mentions for instances and scripts, and a Continue button when the agent hits its tool-call limit. See Chat & Streaming.
@ pulls real place context — instances, scripts, skills — straight into your message, and / runs a command instead of sending plain text. See Mentions & Commands.
A catalog of tools — reading the hierarchy, scripts, and properties; editing scripts with a real diff view; creating, moving, and deleting instances; searching scripts; reading the console; and a locked-down execute_luau escape hatch. See Tools & Permissions.
One scrolling screen: account and billing, Budget Mode, cost info, default model, tool permissions, and Skills/Connectors/Memory nav rows, ending in a double-confirm Danger Zone. See Settings.
Flip it on and the agent starts from a single tool — search_tools — and has to look up anything else by name before it can call it, cutting the fixed cost of a big tool catalog on every turn. See Budget Mode.
Switch between every model your Tomdacat AI account can reach, see which ones support vision or reasoning at a glance, watch reasoning-capable models think out loud in real time, and dial reasoning effort up or down per chat. See Models & Reasoning.
Bring your own per-credit API key, or subscribe to a flat, Robux-billed Coding Plan — switch between them any time in Settings, with usage tracked as session, weekly, and premium-model windows. See Billing.
Watch a live session — connection status and a real transcript — from ai.tomdacat.com/dashboard/code, send it messages from there, and optionally let the dashboard approve tool calls right alongside Studio itself. See Remote Control.
Attach a screenshot from disk and ask the agent about it — gated cleanly to models that actually support vision. See Vision & Images.
Persistent Project Memory you write once, plus Auto Memory the agent writes for itself across sessions — both injected into every system prompt for this place. See Memory.
Normal asks per tool, Plan proposes a plan first, Auto skips every prompt (with an explicit confirmation to turn it on). See Modes.
Delegate a task to a background agent with its own model and tool catalog, tracked in its own rail tab with an unread dot the moment one finishes. See Subagents.
Five built-in playbooks (Luau scripting, DataStores, UI design, 3D building, game systems) plus your own imported skills, referenced with @skill:name or the /skill command. See Skills.
Optional capabilities — like the Fluency Icons connector for search_icons — that you turn on per-project in Settings. See Connectors.

​
Get started

Installation

Install the v3.3 local plugin and connect your Tomdacat AI API key or Coding Plan.

Quickstart

Send your first message and approve your first tool call.