Wren operates as an autonomous agent — not just a response generator. Given a high-level task, Wren will independently plan a series of tool calls, execute them in sequence, adapt when something doesn't go as expected, and continue until the task is done.
This means you can describe complex multi-step goals — "Build a currency economy with purchasable upgrades and leaderboard" — and Wren will break it into sub-tasks, execute them in a logical order, and report back with a cohesive result.
Every proposed change surfaces as a contextual diff. You see exactly what's being added, removed, or changed — line by line, in the same format developers already understand. No black boxes.
The diff viewer shows file-level changes grouped logically. Approve all at once, or step through each hunk. Reject a change and Wren can explain its reasoning or propose an alternative based on your feedback.
Wren doesn't just suggest code in a chat window — it writes directly to your Roblox Studio workspace via the plugin's file tool layer. Once you approve a diff, changes are applied to actual Lua scripts in your game hierarchy.
Wren can create new scripts, modify existing ones, and organize them into the correct service locations (ServerScriptService, StarterGui, ReplicatedStorage, etc.) — all while understanding what already exists and where.
ScriptWEAVER routes tasks to different model tiers based on your selection — Speed for fast turnaround, Balanced for everyday feature work, Quality for deep architecture and complex reasoning.
This isn't a gimmick. Different tasks genuinely benefit from different models. Competitors who lock you to one model are asking you to pay for peak-quality compute on tasks that don't need it — or to accept lower quality on tasks that do.
Full Models Breakdown →ScriptWEAVER is currently designed for solo developers. Team collaboration features — shared sessions, per-contributor history, change attribution, and team model quotas — are on our roadmap.
We're designing this carefully to avoid the "too many cooks" problem. Our approach will be session-isolated: each teammate runs their own Wren session, and changes are reviewed and merged explicitly. More details when we get closer to shipping.
Want to be notified when team features land? Sign up for the mailing list.
Stay Updated →From your first message to files approved and applied — five steps.
Install the ScriptWEAVER plugin from Roblox Studio's plugin marketplace. Connect your account. The panel docks to your Studio layout like any other tool.
Type what you want — a new feature, a bugfix, a refactor. No special syntax required. Switch to Agent mode for autonomous multi-step execution, or Chat mode for back-and-forth assistance.
Using read and list tools, Wren surveys your existing scripts before writing anything. It tells you the plan and you can redirect before execution begins.
You watch the live tool call feed in the Agent panel. Each write operation is staged — not applied yet. You can stop anytime. Wren reports progress at each step.
All staged changes surface as a combined diff. Review file by file. Approve individually or all at once. Rejected changes prompt Wren to revise with your feedback. Approved changes apply instantly to your game.
ScriptWEAVER is built on a server-side agent loop that coordinates tool calls, manages session state, and communicates with AI models via a secure backend. Here's what that means in practice.
Wren interacts with your workspace through a defined set of tools:
list_files, read_file, write_file,
modify_file, search_in_files.
Every workspace interaction is auditable.
Agent jobs track all file reads and writes in a single session graph. Cross-file references (RemoteEvents, ModuleScripts, shared tables) are resolved before any writes are proposed.
Each agent run is persisted as a job object: prompt, tool call chain, diff output, model used, timestamps. This is what populates the History and Logs panels.
Diffs are generated server-side from the original and proposed file contents,
then streamed to the plugin for display. You see a unified diff identical in format
to what you'd see from git diff.
Your model tier selection routes to the appropriate backend model. The agent loop, tool dispatch, and diff pipeline are model-agnostic — tier switching doesn't change how Wren operates, only the reasoning engine.
Server-side loop detection catches runaway agent patterns (e.g., calling the same tool repeatedly without progress). A corrective message is injected and the job is guided back on track, or gracefully stopped.