All your projects. One bench.
TeamAICoder is a desktop workbench for running lots of software projects at once, with AI coding agents doing the building and you doing the steering.
Bring your own agent: Claude Code, Codex, or any command-line tool.
Seven stations, every project.
Each project gets the same seven sections, from the first scribble to running in production. Each section has its own folder, file tree and tabs, so ideas, plans and deploy notes stop ending up in one pile.
Ideas
Notes, pitches and half-formed thoughts.
Design
Mock-ups and specs beside a live preview.
Plan & Track
A status board that shows what needs you.
Build
Agents in terminals, code and docs alongside.
Test
Test runs and CI in tabs next to the code.
Deploy
Hosting dashboards and release notes.
Operate
What's running, and what it costs (coming).
Your agent on the left. The result on the right.
TeamAICoder doesn't replace your coding agent. It gives it a proper workshop: a terminal to run in, the files and web pages it's working on beside it, and a memory of where you left off.
Add your projects
Point it at a folder, or import a portfolio of markdown files. It finds each project's repos on your disk.
Open Chat + Preview
A terminal on the left, a live preview on the right. Start claude, codex or anything else.
Watch it land
Files update live as the agent edits them. Pages the agent opens show in the preview, not a separate browser.
Pick up tomorrow
Relaunch and every tab comes back where it was, including claude --continue in the right folder.
- Electron, context-isolated
- Real PTYs: node-pty + xterm.js
- Browser tabs with persistent sign-in
- Markdown files as the source of truth
- CodeMirror + Toast UI editors
- No account, no server, no telemetry
Small tools, well connected.
Real terminals
Proper PTYs in any folder. Run agents, tests, servers and scripts side by side.
Chat + Preview
Agent and preview in one tab with a draggable divider. Ask for an edit to a SharePoint or Google Docs file and watch it change.
Embedded browser
GitHub, CI, analytics and docs live in tabs. Sign in once, as in a normal browser, and stay signed in.
Markdown first
WYSIWYG or source editing, links between files that work, and live reload when an agent edits the file.
Status boards
Pin a status page to a section. Answer the questions waiting on you, or open Claude on an item, from the board itself.
Remembers everything
Tabs, terminals and previews come back after a relaunch, in every section, for every project.
Every project, every agent, one glance.
The desktop app is where you work. The dashboard is where you check in: a self-hosted server that watches the whole portfolio and runs coding agents on your own Kubernetes cluster while your laptop is closed.
- 1
A portfolio board
Refreshed three times a day from GitHub and your project files: phase, health, CI, open PRs and issues for every repo. You write the intent (phase, blockers, next step) in markdown, and the dashboard observes everything else.
- 2
Signals, not noise
It flags what actually needs you: a project that says it's building but hasn't had a commit in three weeks, CI that's red, or a pull request that's gone stale. The board and the daily digest use the same rules, so they never disagree.
- 3
Agents on your cluster
Start a Claude Code run against any repo. It works on its own branch in its own container, streams its progress live, and stops to ask when it's stuck. Reply from the dashboard and it carries on. It opens a pull request only when you say so.
- 4
Safe by construction
The agent can commit but can't push: only the runner holds the GitHub token. Sessions survive restarts, and stopping a run pushes whatever was committed first, so no work is lost.
- 5
Ask it anything
A daily digest, plus a chat bot with read-only access to the board. Ask "what needs me today?" from your phone.
The dashboard runs privately today. Its source opens up as it becomes the workbench server that the desktop app, a mobile app and an MCP server all talk to.
How we actually ship with it.
TeamAICoder is built with TeamAICoder. Each lifecycle section has a set of Claude Code skills that do the routine parts, and memory carries context from one session to the next.
Pressure-test it
Talk an idea through before writing code: who it's for, what it replaces, the smallest version worth building.
Write it down
A spec, architecture diagrams and an engineering review, all as markdown in git, plus release goals the board can track.
Agent on the left
Claude Code in Chat + Preview, working on a branch. Systematic debugging when something breaks, and the app open in the preview to check.
A second pair of eyes
QA in a real browser, a security pass, a risk rating, and a review by a different model from the one that wrote the code.
Ship through a PR
Every change goes through a pull request with tests and docs checked. Merge only when CI is green and every comment is answered.
Answer what's waiting
The status board pins to Plan & Track: release progress per project, plus a queue of questions and decisions to answer in place.
What's a skill?
A folder with a markdown file of instructions (and sometimes scripts) that Claude Code loads when a task matches, or when you type its name. Skills turn "how we do PRs here" into something the agent does the same way every time, in every project.
# skills live in one folder, shared by every project ~/.claude/skills/ raise-pr/SKILL.md release-board/SKILL.md … # then, in any TeamAICoder terminal $ claude → /raise-pr
Early, useful, moving fast.
TeamAICoder is used every day on macOS. Windows builds are a preview. Here's where it's heading.
Now shipping
- Lifecycle sections, terminals, browser tabs
- Chat + Preview with persistent sign-in
- Projects, session restore, status boards
- Signed, notarised installers
Next building
- Coding agents running on your own Kubernetes cluster
- One workbench API, plus an MCP server
- Portfolio view and an Operate section with costs
Later planned
- A mobile app: an inbox of agents waiting on you
- Push notifications when an agent gets stuck
Get TeamAICoder.
Free and MIT licensed. You'll need your own coding agent, such as Claude Code or Codex, installed separately.
macOS
SupportedmacOS 12 or later. The main platform: everything on this page works here.
Windows
PreviewWindows 10 or later, x64 or Arm. Terminals, editors and browser tabs work. Some extras, such as opening pages from a terminal, are macOS-only for now.
All releases are on GitHub.