Worktree sessions
Run in the project folder when changes should land there, or in an isolated checkout and branch when they should not. Worktree sessions run side by side without touching the same files.
Code Station
FOR macOS
Your agents just got
a headquarters.
Run Codex and Claude Code across every project - conversation, changes, files, and terminal in one window.
Codex and Claude Code.
No favorites.
Pick the agent, model, and effort per session, then run them in parallel, each in its own Git worktree.
Come back.
Caught up.
Ask for a short recap at any time, or turn on automatic recaps for work that finishes while you are elsewhere. The summary stays outside the transcript and keeps only the goal, result or blocker, and next step.
Talk it into shape.
On a live canvas.
Enable the experimental Design mode for a screen, flow, diagram, deck, or prototype. The agent studies the real project, builds beside the conversation, and keeps refining the same interactive artifact.
Three repos.
One conversation.
Group projects into a workspace. The lead is the working directory and the rest join the same conversation.
Ship the endpoint.
Then poke it.
A built-in HTTP client with staging and production environments, OAuth in the Keychain, and {{env}} that resolves itself.
Bad deploy?
Send in the agent.
Guided troubleshooting with skills, MCP servers, and read-only guard rails when the environment is live.
And the rest of
the window.
The parts you reach for between prompts, in the same place as the conversation.
Review and commit
Pick exactly the files that go into a commit, amend the last one while it is still unpushed, and read recent commits with their full diffs.
Code that reads as code
Chat code blocks, file previews, and diffs are syntax highlighted, with no extra dependencies.
Rewind and fork
Take a Claude Code conversation back to an earlier prompt and send it again, or fork it into a new session from that point.
Cost and context
Every session shows its spend next to the context meter. Any agent's settings can hide the figure.
Notifications
A system notification when a session finishes a turn or stops to ask something. Clicking it opens that session.
MCP servers
Configure servers once and register them with Codex, Claude Code, or both. Tag each one with the environment it belongs to.
Skills
Install and update agent skills from your organisation's marketplace.
Shortcuts and Docker
Save, run, and stop shell commands from the status row, and inspect or stop running containers without leaving the app.
Install it.
On your Mac.
Open the latest release on GitHub and download the signed and notarized DMG. Open it, drag Teya Code Station to Applications, and launch the app - no toolchain, no build step.
What you need
- macOS 14 or later
- At least one agent installed and signed in:
codexorclaude
For contributors
Building from source needs Xcode 16 (or another Swift 6 toolchain) and Git. The script makes a double-clickable bundle you can drop in Applications.
$ git clone https://github.com/teya-engineering/code-station.git
$ cd code-station
$ ./build-app.sh
$ open "build/Teya Code Station.app"Make it yours
The parts that belong to your organisation rather than to the app - the identity provider your APIs sign in against, your environments, your Grafana instances, the skills marketplace, the commands worth having on hand - all live in one optional settings file the app reads at startup. Every section is optional, and so is the file.
The first-launch wizard can read it from a local file or clone it from a Git repository, and build-app.sh can fold it into the bundle so the app you hand to your team is already set up. See the site configuration guide and the blank-slate example.