Skip to content

Every project. Every agent. Every key. One window.

Styx is where everything converges: every repo you work in, the AI coding agents building in them, and every login they need to ship. Switch between all of it from one window. And like the river it's named after, nothing crosses over without you. Deploying, or touching live data, asks you first.

Free and open source under Apache-2.0. Read the code on GitHub.

Styx - An agentic development environment (ADE) for Mac | Product Hunt

An agentic development environment (ADE) for Mac, with Windows and Linux in beta. Your keys never leave your computer.

Codex tries to push a change to the live database. Styx catches it before it goes anywhere.

00Needs you
03Agents working
02Grants active
05Projects

How it works

Every project

All your repos in one window, each with its own agents, branches and services. A glance shows what needs you; one keystroke takes you there. No more hunting through windows and terminals.

Every agent

Claude Code, Codex, Gemini CLI, Cursor and OpenCode, side by side, each on its own copy of your code. Read what they did, change by change, keep what you like, or open it in your own editor with one key.

Every key

Log into Vercel, Supabase, AWS, Google Cloud, Fly.io, Railway, Render, GitHub and your servers once, in one place. Agents never hold the keys. They borrow access, for an hour, with your fingerprint for anything live. Every borrow is written down.

What happens at the crossing

Agents build freely on your side of the river. To deploy, or touch live data, they have to cross, and nothing crosses without you. It asks, you decide, it gets a temporary pass, and the record keeps the receipt.

  1. 01

    The agent tries something real

    Say it runs a deploy, or wants to change your live database. Styx catches the request before it goes anywhere. The agent never sees a password or a key, so there is nothing for it to leak or misuse.

  2. 02

    It waits for you

    The agent pauses. You see the request wherever you happen to be: in its chat, on the board, in your inbox, as a notification, as a badge on the dock. Answer in any one place and it’s answered everywhere.

  3. 03

    You decide how much to allow

    Just read, or also write? This once, for an hour, or until the agent is done? Anything that touches production asks for your fingerprint or face before it goes through.

  4. 04

    The agent gets a temporary pass

    It works only for what you allowed, only for that agent, and only for as long as you said. When the time is up, or the agent finishes, or you change your mind, the pass is gone. Nothing permanent is ever handed over.

  5. 05

    It’s written down

    Who asked, what they got, for how long, and what they did with it. The record can’t be edited or deleted, by anyone, ever. Taking a pass away adds a line; it never removes one.

Everything you already use, in one place.

Styx runs the agents you have installed and signs them in for you, connects to the services you are already logged into, and picks up your editor on first launch: recent folders come along, and anything can be opened back in it with one key. Nothing in your editor changes.

Your editor

  • VS CodeRecent folders come along, and anything opens back in VS Code.
  • CursorSame as VS Code: recent folders, and Open in Cursor.
  • WindsurfSame as VS Code: recent folders, and Open in Windsurf.
  • JetBrains IDEsWebStorm, IntelliJ IDEA, PyCharm, GoLand, RustRover. Recent projects come along.
  • ZedDetected and ready under “Open in”.
  • NeovimDetected and ready under “Open in”.

Agents

  • Claude CodeChat with it inside Styx, approve its plans, answer its questions.
  • CodexSame chat, same approvals.
  • Gemini CLISame chat, same approvals.
  • CursorSame chat, same approvals.
  • OpenCodeSame chat, same approvals, with any provider it supports.
  • Your terminalA plain shell, with the same rules about what it can reach.

Things agents have to ask for

  • VercelDeploys and previews.
  • SupabaseYour databases, live and staging.
  • AWSAnything you can reach with the aws command.
  • Google CloudAnything you can reach with gcloud.
  • Fly.ioYour apps, with a token made for each grant.
  • RailwayEach environment, with its own short-lived token.
  • RenderDeploys and services.
  • GitHubRepos, pull requests, releases.
  • Your serversOver SSH, using the identity already on your computer.

Agents don't step on each other. Or on you.

Every agent works on its own copy of your project, on its own branch. You keep working on yours. When something clashes, the agent stops and tells you exactly where.

BranchWhoChangesPull requestAction
mainyou——
fix/checkoutClaude+142 −38 · 3 files#214 openReview
test/flakyCodex+12 −4 · 2 files—Open
docs/readmeGeminiconflict README.md—Resolve
  • Nothing merges on its own. An agent's work stays on its branch until you press Land, which brings in the latest main, runs your checks and merges it in one step.
  • Clashes go back to the agent that made them. It is told what changed underneath it and finishes the merge itself. Agents working on the same file are warned before it becomes a clash.
  • Finished work tidies itself away. Landed branches archive themselves; agents you mark done stay on the board for a week, then go.

Design it first. Then point at what's wrong.

Agents can draw your screens before they write code, and fix what you point at once it runs. Nothing to describe in a paragraph: click it.

Every task, one card.

Start as many tasks as you like, each with its own agent on its own branch. The Tasks tab shows each one as a card: what it is doing, whether it needs you, and what it changed. Click a card to open its chat.

  • Mark a task Design or Build when you start it.
  • A design and the build that came from it point at each other.
  • Your turn shows in lime, so the one that needs you is easy to find.
The Tasks tab: cards for a design task, a build task, a flaky test fix and an idle Gemini lane, each with its agent, branch, status and changes, and a card to start another lane as Design or Build

Design it before you build it.

A design task draws your screens first, as wireframes or finished designs, at desktop, tablet and phone size, in your own type and colours. Every size sits side by side on one canvas.

  • Click anything to change its text, colour or corners, or drag over an area.
  • Or tell the agent: it gets the screen, the element and a picture of it.
  • Build it hands the screens to a build task, and tells it when the design changes.
  • Screens are plain files in your repo, reviewed and kept like code.
The Design tab: a checkout screen at desktop, tablet and phone size, the Pay button selected, and a panel to change its text and colour or tell Codex to make it sticky on phone

Point at what is wrong.

Preview runs your app beside the chat. When something looks off, press Select to fix, click it or drag over an area, and say what is wrong. The agent gets the element, where it lives in your code when the page says, and a picture.

  • Desktop, tablet and phone sizes.
  • A design problem goes to the design task instead.
  • It only shows pages on your own machine. A repo can’t point it anywhere else.
Preview: the running checkout page beside the chat, with Fix this naming the Pay button and a note about its hover contrast, ready to send to Claude

From a finished branch to live, without leaving.

Once an agent is done, everything else happens in the same window too. Anything that reaches production still asks you first.

Straight to GitHub.

No repo yet? Create one, private by default, or link one you already have. Publish then commits, pushes and opens the pull request in one step, with the message drafted by your agent from the changes.

  • It uses the GitHub login you already have on this machine.
  • Files like .env and private keys are kept out of the commit.
  • The push gets its own short-lived access, recorded in the log and thrown away when it finishes.

Deploys the way you would do it yourself.

Each place you deploy to runs the command you’d type: vercel, gcloud, a script over ssh. You don’t have to know it. On the first deploy an agent works it out, and Styx remembers.

  • Anything live asks for your fingerprint first.
  • Progress stays on the button and in the status bar, so you can keep working.
  • Every deploy is in the log: who asked, where it went, what ran and how it ended.

Find what will slow the next person down.

Tech debt audit has an agent read your project the way a new engineer would, and report what will actually trip someone up. It runs in the background and changes nothing.

  • Every finding names the file and line, the mistake it leads to, and the size of the fix.
  • It checks your setup instructions against what the code really needs.
  • Switch projects while it works; the report waits under Tasks.
  • Keep talking while it works.

    Type your next message while an agent is busy. It waits for the current turn to finish, or Codex takes it straight away.

  • Undo a turn.

    Every turn an agent takes is saved. If one goes wrong, put the project back to how it was before it, in one click.

  • Skills for every agent.

    Install a skill once and choose which agents get it: Claude Code, Codex, Gemini CLI, Cursor, or all of them.

Also in the app: a usage view that shows how close each agent is to its limit, and an agent dock that keeps every chat above your other windows.

What Styx will never do.

An oath sworn on the Styx could not be broken, even by a god. These are ours, built into the software rather than written in a policy.

  • Keep your passwords anywhere but your computer’s own secure storage.

    Keys live in your system’s keychain (the Mac Keychain, Windows Credential Manager or the Secret Service keyring on Linux), nowhere else. Not in a file, not in a log, not in the app’s own database.

  • Let anything reach production without you.

    Deploying, or changing live data, asks for Touch ID, Windows Hello or your system password first. Every time.

  • Give an agent a permanent key.

    Every pass is temporary and covers only what you approved. When it runs out, the agent is back to asking.

  • Erase the record.

    Every request and approval is logged in a way that cannot be edited or deleted. Taking a pass away adds a line.

  • Send your work to us.

    Your code, your projects, your prompts and your keys stay on your computer, and your agents talk to your services directly. Styx’s own server knows your account and, unless you switch it off, how often a handful of features get used. Never what you used them on.

Fast, if you want it to be.

Everything has a key. The command bar is where you start when you don't know which one.

  • ⌘KOpen the command bar: switch, start an agent, deploy, approve
  • ⌘PSwitch project
  • ⌘1..4Jump to an agent's chat
  • ⌘⇧NStart a new agent
  • ⌘⏎Approve the request in front of you
  • ⌘⌫Turn it down
  • rUndo a change
  • jkNext or previous change
  • ⌘⏎Finish reviewing
  • ⌘⇧OPop a chat out into its own window
  • VSelect on a design or in Preview, to change it or tell the agent
  • ⌥⇧←⌥⇧→Move the current tab left or right
  • ⌘⇧TSwitch light and dark. Works on this page too.
  • escClose whatever is open

Download

Version 0.4.14, free and open source. For Mac, with Windows and Linux in beta.

Mac

Touch ID for anything that touches production, or your Mac's password on one without it. Apple silicon.

Download

Windows

Windows Hello for anything that touches production. Same app, same rules. Windows 10 and 11, x64.

Download

Linux

Your system password for anything that touches production. An AppImage for any distribution, or a .deb for Debian and Ubuntu. x64.

Download on GitHub ↗

Styx works with the agents already installed on your computer, and can install the ones you're missing. Installing Styx covers each platform, including what Linux needs.

Open source.

Styx is free and open source under the Apache License 2.0. Read how it works, build it yourself, or help it reach more agents and more places to deploy.

  • Read the code

    All of it is on GitHub, including the rules that decide what an agent may reach and the audit log that records it. Build it yourself with Node, pnpm and git.

    Styx on GitHub ↗
  • Add an agent

    Styx runs Claude Code, Codex, Gemini CLI, Cursor’s agent and OpenCode. The guide covers teaching it another one, from finding the CLI to the tests a pull request needs.

    Read the guide
  • Add a deploy target

    Vercel, Supabase, AWS, Google Cloud, Fly.io, Railway, Render, GitHub and SSH today. The guide shows how a new provider gets the same asks, time limits and record as the rest.

    Read the guide

Questions, ideas and things you built go in GitHub Discussions. Bugs go in issues.