Install
macOS / Linux:
curl -fsSL https://raw.githubusercontent.com/Lautaro005/workstudio/main/install.sh | sh
With brain, Lumina and OpenCue too:
curl -fsSL https://raw.githubusercontent.com/Lautaro005/workstudio/main/install.sh | sh -s -- --all
Windows (PowerShell):
irm https://raw.githubusercontent.com/Lautaro005/workstudio/main/install.ps1 | iex
On macOS the app goes to /Applications/AI WorkStudio.app (or ~/Applications with --user), on Linux to
~/.local/opt/workstudio with a menu entry, on Windows to your user's programs. All of them add a workstudio
command: workstudio opens it, workstudio update updates it, workstudio uninstall removes it (your settings in
~/.workstudio stay). No release yet for your platform? Add --build to build from source (needs git, Node and Rust).
What it does
| |
|---|
| Workspace | Home with every app's status and version; sidebar with live status; Google-style app launcher; ⌘K palette. Spanish and English. |
| Apps inside | Each app opens on its details; brain's dashboard and Lumina can also be used inside WorkStudio (with the preferences of their own window), or open there directly if you check it. OpenCue opens in its own window. |
| Chat | Talk to Claude Code, Codex, Gemini, any Ollama model or an OpenCue agent. Pills turn on brain, Lumina and web search (built into WorkStudio). Conversations stay in ~/.workstudio/chats. |
| Terminals | Up to six real terminals in a grid. Profiles for Claude Code, Codex, Gemini CLI, Cursor Agent, brain, lumina and opencue. |
| One MCP server | workstudio mcp joins the tools of your apps (and the servers you add) under one server, named <app>__<tool>. Terminals and the chat get it automatically; connect Claude Code, Claude Desktop, Codex, Cursor, Gemini or Windsurf with one click (with a backup). |
| Notifications | brain vault changes, Lumina meetings, apps that start or stop, new versions — in the app and as system notifications. |
| Updates | Checks GitHub releases and npm every few hours; updates each app and CLI from a terminal and reads the installed version again when the command ends; WorkStudio updates itself (signed bundles). |
| Models | Settings → Models: choose which models of each provider show in the chat picker and add your own by ID. |
One MCP server
workstudio mcp is an MCP server that starts the servers of your apps, merges their tools and routes each call:
brain__search_knowledge, lumina__…, opencue__…. Agents see one server instead of a long list.
workstudio mcp --only brain,lumina,web limits it; web adds WorkStudio's own web__search and web__fetch.
WorkStudio writes it to ~/.workstudio/mcp/claude.json and to a workspace folder (~/.workstudio/workspace) that
agent terminals start in:
- Claude Code →
claude --mcp-config ~/.workstudio/mcp/claude.json
- Codex →
codex -c mcp_servers.workstudio.command=… -c mcp_servers.workstudio.args=[…] (your config.toml is untouched)
- Gemini CLI →
.gemini/settings.json in the workspace
- Cursor Agent →
.cursor/mcp.json in the workspace; VS Code → .vscode/mcp.json
HTTP servers you add are passed next to it, as they are. The servers it carries:
| App | Server |
|---|
| brain | uv run --directory ~/.brain python server.py |
| Lumina | lumina mcp |
| OpenCue | opencue mcp-serve --agent <id> (pick the agent in MCP) |
The chat
| Model | How it runs |
|---|
| Claude Code (Opus / Sonnet / Haiku latest and each version: Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1, Opus 5 … Haiku 4.5) | claude -p headless with your login, only the workstudio server (--strict-mcp-config) and no built-in tools |
Codex (default, a latest per family and every version: your account's models from ~/.codex/models_cache.json first, then the GPT versions Codex knows) | codex exec --json, read-only sandbox, workstudio passed with -c |
| Gemini CLI | gemini -p in a folder whose .gemini/settings.json has only workstudio |
| Ollama | /api/chat from WorkStudio, which runs the tool calls itself |
| OpenCue agent | OpenCue's send_to_agent: the agent answers with its own tools (the pills don't apply) |
Adding an app
New apps of the family go in catalog/apps.json (shipped inside WorkStudio). For one of your
own, add an entry like this to the apps list of ~/.workstudio/apps.json; it wins by id:
{
"id": "my-app",
"name": "My App",
"tagline": "What it does",
"kind": "web",
"url": "http://127.0.0.1:9000/",
"health": { "url": "http://127.0.0.1:9000/api/health", "contains": "ok" },
"cli": "my-app",
"repo": "me/my-app",
"start": { "unix": "my-app serve", "windows": "my-app serve" },
"install": { "unix": "curl -fsSL https://example.com/install.sh | sh", "windows": "irm https://example.com/install.ps1 | iex" },
"update": { "unix": "my-app update", "windows": "my-app update" },
"mcp": { "command": "{cli}", "args": ["mcp"] }
}
Placeholders in MCP commands: {cli} (full path of the app's command), {home} (its home folder), {uv}, {ws}
(~/.workstudio).
Data
Everything WorkStudio keeps is in ~/.workstudio (WORKSTUDIO_HOME overrides it): settings.json, apps.json,
mcp.json, notifications.json, mcp/ (generated configs), workspace/ and logs/.
Development
cd apps/desktop
npm install
npm run dev # UI in the browser with a mock backend (http://localhost:1430)
npx tauri dev # the desktop app
npx tauri build # bundles (set TAURI_SIGNING_PRIVATE_KEY for updater artifacts)
Rust checks (what CI runs): cargo fmt --check, cargo clippy --all-targets -- -D warnings, cargo test in
apps/desktop/src-tauri.
Stack: Tauri 2 + Rust (portable-pty, ureq, toml_edit) and React 19 + TypeScript + zustand + xterm.js. Same family
pattern as brain, OpenCue and
Lumina.
Releasing
Bump the version in apps/desktop/package.json, src-tauri/Cargo.toml and src-tauri/tauri.conf.json, add a
CHANGELOG.md entry, then push a tag vX.Y.Z. The Release workflow builds macOS (universal), Linux and Windows
installers plus latest.json for the updater. It needs the repository secret TAURI_SIGNING_PRIVATE_KEY (the key
whose public half is in tauri.conf.json).
License
Apache-2.0 with the Commons Clause — free to use, modify and share, not to sell. See LICENSE and
SECURITY.md.
