Desktop app for macOS & Windows
All your AI coders. One workbench.
Build real software with the Codex and Claude Code access you already have. Add Cursor, OpenRouter or local models in LM Studio. Work in your own folders, and decide who leads and what they may touch.
Works with
01Providers
Plug in what you already pay for.
Connect one route or all five. Each keeps its own sign-in, and usage is billed by that provider, not by Mythra Code. Tap a socket to patch it in.
Mythra Code
2 routes live
02Crews
Mix models in one thread.
Put GPT Sol in charge and let it hand part of the job to Claude Opus, or to any route you approve. Crews stay off until you turn them on. Try it.
Who leads
Crew
Task
“Add a dark mode to the dashboard”
A toy simulation of the controls, not a recording.
- Off by default. Nothing delegates until you switch a crew on.
- Only what you approve. Pick each destination and allow up to 24 at once. Each one uses its own provider's plan.
- Watch it, stop it. Open any sub-agent to see its work, or stop just that one.
03Control
You stay in charge.
Every thread runs in one of three modes, set right under the message box. Try them.
- Read and explain your code
- Edit files in the project
- Elevated actions, like network access
- Your prompt starts blank. App-wide instructions are empty until you write them. Each project can inherit, replace or add its own.
- Roll changes back. Checkpoints snapshot your code before and after each turn. Isolated worktrees keep risky work apart.
- No telemetry. Chats, settings and usage stay on your computer. Prompts go to the providers you choose.
04FAQ
Questions, answered.
Is Mythra Code an AI model?
No. It's a desktop workbench around the coding tools you connect. The models come from your providers. Mythra Code handles the project, permissions, sub-agents and everything around the conversation.
Can I use my ChatGPT or Claude subscription?
Yes, through each provider's official route. For OpenAI, you sign in through the Codex runtime with a ChatGPT account that includes Codex. For Claude, Mythra Code drives your installed Claude Code CLI and never asks for your password. Your plan's limits still apply.
Do I need all five providers?
No, connect only what you want. The OpenAI, OpenRouter and LM Studio routes need the Codex runtime, which isn't bundled.
Can I run local models?
Yes, through an LM Studio server on your machine or one you trust. Whether tools work depends on the model you choose.
Which computers does it run on?
macOS on Apple silicon, and Windows x64. Intel Macs, Linux, phones and tablets aren't supported.
Can I undo what the agent did?
Usually. Checkpoints snapshot your code before and after each turn, and restoring takes a safety snapshot first. Worktrees and built-in Git help too. Files ignored by .gitignore aren't covered.
Does it cost extra?
Model usage is billed by the provider you pick, under your own plan or API key. Running several sub-agents at once uses that allowance faster.
What leaves my computer?
Your prompts go to the providers you choose. A local LM Studio server keeps them on your machine. The app also checks for updates and model catalogs, and runs any GitHub actions you start. The app has no telemetry, analytics or crash reporting.
Why does Windows say “Unknown publisher”?
The Windows installer isn't Authenticode-signed, so Windows can't show a publisher name. Download from this website or the official GitHub release. In-app updates are signature-verified separately.
05Download
Get Mythra Code.
- Download and install the version for your computer.
- Connect a provider in Settings → Models & accounts.
- Open a folder, pick a model, and describe the task.