Portus — Central de Controle para Agentes IA
adospace/portus
Controle múltiplos agentes IA em uma janela — uma central de missão para CLIs.
Ver no GitHubAplicação desktop multiplataforma que funciona como central de controle para múltiplas sessões de agentes IA CLI (Claude Code e outros). Permite executar, rastrear e alternar entre várias sessões de terminal em uma única janela com três painéis, monitorando status, contexto e histórico de todas as sessões simultaneamente.
Instalação
npx convez-ai adospace/portus1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Executar simultaneamente múltiplas sessões de Claude Code em diferentes diretórios
- Monitorar status de ocupado/ocioso/concluído de vários agentes sem alternar abas
- Rastrear consumo de contexto de cada sessão e gerenciar histórico persistente
- Gerenciar fluxo de trabalho complexo com múltiplos agentes IA em paralelo
Veredito

Portus é uma ferramenta sólida para quem trabalha intensamente com agentes IA CLI. Resolve genuinamente o problema de gerenciar múltiplas sessões simultâneas, oferecendo visibilidade em tempo real e persistência entre reinicializações — essencial para fluxos de trabalho complexos.
README
A tradução para o português deste README ainda está na fila.
portus — Agentic Terminal
Mission control for multiple AI agent CLI sessions. Run every agent in one focused, three-pane window and watch them all at a glance.
What is portus?
portus is a standalone cross-platform desktop app (Windows + macOS) that acts as mission control for multiple AI agent CLI sessions — Claude Code today, Codex and other CLIs later. Instead of juggling several terminal windows, you run every agent in one focused, three-pane interface.
It is deliberately not an IDE: there is no code editor, no file viewer, no git integration. The app does one thing well — spawn, track, and switch between long-running agent terminals — and gets out of the way.
Why use it over a plain terminal
- Many agents, one window. Each session is a tab with its own real PTY (a true shell + agent CLI, not a fake terminal), so anything that works in your terminal works here.
- At-a-glance status. Every tab shows whether its agent is busy (●, pulsing), idle (◌), or done (✓), plus a live elapsed timer while it's working — so you know which sessions need attention without clicking into them.
- Start sessions where the work is. The left pane is a lazy folder tree rooted at a drive/volume you pick; right-click any directory to launch an agent there or open it in your OS file manager. Switching tabs highlights the active session's folder and its parents in the tree, so you always see where you are.
- Context-window gauge. Each tab shows how full its session's context window is —
a thin fill bar along the bottom edge (green → amber ≥80% → red ≥95%) plus a
% ctxreadout in the status bar — so you know when to/compactor start a fresh session. - Persistence. Sessions (working folder, agent session id, title) survive app restarts and reappear in the session-history pane.
- Searchable session history. The right pane lists every session by the title the agent CLI generates for its task (falling back to the folder name) plus the working path; a search box filters by title, name, or path, and one click restores a session. Hover a row to open its folder in the file manager or delete it from history.
- Flexible layouts. Split the center into 1, 4 (2×2), or 6 (3×2) panes, each its own tab group, and drag the gutters to resize. Right-click a tab to relocate it between panes, open its folder, or close it.
- Light & dark themes. Follow the OS, or force light/dark — set it in Settings ▸ General, where you can also pick the default folder for new sessions and how long session history is kept.
┌─────────────┬──────────────────────────────┬─────────────────┐
│ Folder Tree │ Terminal Tabs │ Session History │
│ │ [● proj-a] [◌ proj-b] [+] │ │
│ ~/dev │ ┌──────────────────────────┐ │ proj-a 14:32 │
│ ├ proj-a │ │ │ │ proj-b 13:10 │
│ ├ proj-b │ │ xterm.js canvas │ │ ... │
│ └ proj-c │ │ │ │ │
│ │ └──────────────────────────┘ │ │
│ │ 📁 ~/dev/proj-a 63% ctx │ │
└─────────────┴──────────────────────────────┴─────────────────┘
Install
Grab the latest installer from the Releases page.
Windows
Download portus-Setup-<version>.exe and run it. The NSIS installer lets you choose
the install location and creates Start Menu / desktop shortcuts.
macOS (Apple Silicon)
Download portus-<version>-arm64.dmg, open it, and drag portus to Applications.
The macOS build is not yet code-signed or notarized, so Gatekeeper will warn that the app is from an unidentified developer. To open it the first time: right-click the app → Open → Open, or run
xattr -dr com.apple.quarantine /Applications/portus.app.
Note: macOS builds are Apple Silicon (arm64) only.
Prerequisites
portus is a launcher for agent CLIs — it spawns them in a real shell. To actually use
it, you need the agent CLI installed and on your PATH, e.g.
Claude Code (claude).
Build from source
Requires Node.js 20+ and npm.
npm install # node-pty ships N-API prebuilds — no native rebuild needed
npm start # build + launch Electron
Package distributables
npm run dist:win # → release/portus-Setup-<version>.exe (NSIS installer)
npm run dist:mac # → release/portus-<version>-arm64.dmg + .zip
Each dist:* script builds the renderer/main bundles before running
electron-builder. Output lands in release/.
Releasing
CI (.github/workflows/release.yml) builds Windows + macOS and publishes the
installers to GitHub Releases whenever a v* tag is pushed:
npm version patch # bumps package.json and creates a vX.Y.Z tag
git push --follow-tags
Architecture
A single Electron app. The main process is the sole broker — it owns the UI window,
node-pty PTY management, and session persistence; the renderer reaches privileged
work only through the window.api contextBridge, never node-pty or the filesystem
directly. Sessions are persisted in-process as a small JSON file in the per-user app
data dir (no database, no second runtime); context-window occupancy is read live from
each session's Claude Code transcript.
Built with Electron, TypeScript (strict), xterm.js, node-pty, and Tailwind CSS v4.
See CLAUDE.md for the full design and file map.
License
MIT