⇡
Zero inbound ports
StudioCat polls OpenAI over HTTPS. No local server, no port forwarding, nothing exposed to your network.
Connect AI Chat
to Your Local Workspace
Your AI chat reads, edits, runs and rolls back code on your Mac,
with every change visible.
Free during the beta · Works with ChatGPT on the web
AI chat · web
The
Add result in the calc package is wrong. Fix it and run the tests.
Found it —
Add subtracts. I changed one line and ran
go test ./....
✓ All tests pass.
Ask anything
ActivityReview DiffOutput
- ✓
open_workspace - ✓
grep_search - ✓
apply_patch - ✓
run_command
calc.go
- return a - b + return a + b
How it works
Three steps, then just chat.
-
1
Create a secure tunnel
In OpenAI Platform, create a tunnel and a restricted key with Tunnels Read + Use. No model API billing is involved.
-
2
Paste it into StudioCat
The setup assistant stores it in your Keychain and connects. Your Mac only dials out — nothing listens.
-
3
Add the MCP server in ChatGPT
Choose Connection: Tunnel, pick your tunnel, start a chat. Ask it to fix, refactor, run and review.
ChatGPT
on the web
on the web
tool call
OpenAI
Secure Tunnel
Secure Tunnel
outbound only
StudioCat
on your Mac
on your Mac
Features
A local workbench for the AI you already use.
⛨
Per-call authorization
Every call is checked by tool, command and path against the projects you approved. Terminal and commits are off by default.
◫
Every change visible
The workbench shows each tool call, patch and command output as it happens — not after the fact.
↺
Roll back any patch
Each edit lands as a card in Review Diff. Undo it with one click, including renames and file modes.
⎇
Checkpoints that stay out of your way
Snapshots live in shadow refs. Your branch, HEAD, index and stash are never touched.
⧉
Parallel projects
Each conversation stays bound to its own project. Switch what you look at; work keeps running where it belongs.
{ }
Native code intelligence
Diagnostics, definitions, symbols and outlines from real language servers — Swift, Go, TypeScript, Rust and more.
?!
Honest results
If a call times out, the model hears “outcome unknown — check first”, never a false “nothing happened”.
Security
Built for code you care about.
StudioCat runs commands on your machine, so the defaults are conservative and the boundaries are enforced on every call — not suggested to the model.
- Credentials stay in your Keychain. Never in arguments, logs or status files.
- Only folders you approve. Paths are resolved and checked before anything runs.
- Dangerous commands ask first. And a denial is final — the call has no side effects.
- Local audit log. Every allow and deny is recorded on your Mac.
- No telemetry by default. Nothing is sent anywhere unless you opt in.
- Plain about limits. Checkpoints are not backups, and anything the model reads becomes part of your chat.
FAQ
Questions
Which AI apps does it work with?
ChatGPT on the web, in a desktop browser, using developer mode MCP servers. The ChatGPT mobile apps can't call custom MCP servers yet. More AI providers are planned.
What does it cost?
StudioCat is free during the beta. It uses your existing ChatGPT account — there is no model API billing. You need an OpenAI Platform organization to create the tunnel.
Does my code leave my Mac?
StudioCat never uploads your repository. The model only sees what it reads through tool calls — those contents become part of your ChatGPT conversation, like anything you paste.
What do I need?
A Mac with macOS 14 or later, a ChatGPT account with developer mode, and the toolchains your projects already use.
Is it open source?
Not yet. Signed and notarized builds come first, together with a public security design note.
The private beta opens soon.
Signed, notarized builds with automatic updates. Leave your email and we'll send the download link.
Request beta access