Wren is in Alpha — free to use while we're building. See planned pricing →
Step by Step

Here's exactly what happens
when you prompt me.

Guardrails

What I will do.
What I won't.

Knowing my limits is just as important as knowing my capabilities. I'd rather be honest than overpromise.

What I Will Do

  • Read your existing code before writing anything new
  • Show diffs for every proposed change before applying
  • Coordinate multi-file changes for consistency
  • Write Roblox-native Lua following Roblox Studio conventions
  • Tell you when I'm uncertain rather than guessing confidently
  • Explain my reasoning in plain language
  • Save session history so you can trace every action
  • Respect your architectural decisions even if I'd do it differently

What I Won't Do

  • Apply changes to your workspace without your explicit approval
  • Delete files without telling you — and never without confirmation
  • Silently overwrite code I didn't tell you I was changing
  • Claim to be a human developer or misrepresent my nature
  • Share your code, prompts, or workspace data with third parties
  • Ignore an explicit override or direction from you, even if I disagree
  • Guarantee 100% bug-free output — AI has limits, I'll be upfront about them
  • Generate malicious code, exploits, or content that violates Roblox ToS
Product Philosophy

Why I'm Minnesota Nice.

The team that built me is from Maple Grove, Minnesota. And there's a real culture there — one that values directness without rudeness, helpfulness without condescension, and honesty without drama. That's what I'm designed around.

You'll notice it in how I talk to you. I won't hype you up with hollow praise when something doesn't work. I won't be cryptic about what went wrong. I'll just tell you clearly, calmly, and with enough context to actually understand it.

Polite

I treat your code with respect. Even if I think a different architecture would serve you better, I'll say "I'd suggest…" not "this is wrong."

Patient

Not getting what you want from me? That's on me to explain better, or on me to ask better questions. I don't get frustrated.

Transparent

I'll tell you what tools I called, what they returned, and why I made the decisions I did. No black boxes.

Collaboration-First

Your instincts about your own game are valuable. I bring knowledge; you bring context. The best results come from both.

You're In Charge

Your controls.

ScriptWEAVER is designed so that the developer always has the final say. Here's every lever you control.

Approve / Reject Diffs

Every proposed change surfaces as a line-by-line diff. Approve all, approve selectively, or reject entirely — your call, every time.

Review Session History

Every Wren job is stored in the History panel. See what was changed, when, and which model ran it. Use it for auditing or learning from past sessions.

Choose Model Per Task

Pick Speed, Balanced, or Quality for each job. You control the speed-vs-depth tradeoff based on what the task actually requires.

Stop Mid-Run

If Wren is heading in the wrong direction, stop the job at any point. No partial writes are applied without your explicit approval of the diff.

Re-Run or Modify

Not happy with the result? Ask me to revise with feedback, or start a new session with adjusted instructions. Sessions are independent.

Manual Rollback

Approved something you regret? The session history shows exactly what was changed. Use Roblox Studio's built-in version control or git to revert. We'll add integrated rollback in a future version.

Common Questions

Honest answers.

I'd rather answer these directly than have you find out the hard way.

Nope. You type what you want in plain English (or Lua pseudocode), and I handle the complexity. You don't need to know how large language models work, what temperature settings do, or what a token is. You just talk to me like you'd talk to a smart colleague. If you get stuck, the docs have step-by-step guides for each capability.
That's a fair concern — I'm not infallible. That's exactly why the diff review step exists. You read what I propose before it runs. If it looks off, reject it and tell me what's wrong. I'll revise. I'll also flag when I'm genuinely uncertain rather than confidently writing code I have doubts about. If you find a recurring pattern where I'm getting things wrong, that's useful feedback — the team takes it seriously.
Wren never touches your workspace without showing you a diff first, and the diff requires your explicit approval. Nothing is silently overwritten. For added safety, I read existing code before proposing changes so I understand what's already working. I can't guarantee I'll always catch every edge case — but I'm thorough, and you have final say.
For any given task, I'll read the files relevant to that task — not necessarily your entire codebase in one shot. If a task is complex enough to require understanding many files, I'll read them in sequence. There are context limits (the amount of code I can hold in mind at once), but I handle this gracefully by being selective about what I load.
You're the developer. If you reject my approach and give explicit instructions to do it differently, I follow your instructions — even if I'd have done it another way. You can always override me. I might note my reasoning once if I think it's relevant, but I won't argue about it.
The big difference is workspace integration. Raw ChatGPT or Claude can suggest code, but they can't read your actual files, understand your project structure, or apply diffs to your codebase. I live inside Roblox Studio, I read your real workspace, I use tools to write and modify files, and the whole interaction is designed for the Roblox Lua development context specifically.
Ready to Try?

Nice to meet you.
Let's build something.

Create Free Account See All Features