Skip to content
Projets

puritysb/AgentDeck has 297 GitHub stars and 59 forks. Star changes are not measured yet. Last checked 2026-11-13 22:54 UTC. Last push 2026-11-13.

Comment ces chiffres sont mesurés

Progression mesurée des étoiles

in 24h
Historique en cours
in 7 days
Historique en cours
in 15 days
Historique en cours
in 1 month
Historique en cours

Variations nettes sur des périodes qui se chevauchent, pas des totaux quotidiens. Les mesures absentes restent vides. Les étoiles indiquent l’intérêt, pas la qualité.

À propos

Local dashboards and physical controls for AI coding agents — native 3D aquariums, Stream Deck, Apple/Android apps, e-ink, ESP32 and LED displays.

Outils de développement

README.md

Original

AgentDeck icon — aquarium dome with octopus and crayfish on a Stream Deck control surface

AgentDeck

App Store — Mac, iPhone, and iPad Google Play — Android Elgato Marketplace Ulanzi Marketplace Vobot Mini Dock app — Gallery 0.3.0 approved MIT License npm version CI Website

Your AI agents, at a glance. Across your desk.

AgentDeck brings your AI agents into one local dashboard. See which are working, waiting for you, or finished—across projects and tools. Watch them in a living aquarium, follow the timeline, and keep usage limits in view. Use your Mac or terminal on its own, or add a tablet, Stream Deck, e-ink reader, or small desk display.

AgentDeck on a real desk, photographed September 28, 2026: iPad and Android dashboards, Stream Deck and Ulanzi controls, e-ink readers, ESP32 panels, and LED matrices

Photographed on September 28, 2026. See the full desk.

Start with what you already have. No extra hardware required. Use your Mac or terminal on its own, then add the displays that fit your desk.

Get started → · Choose your view ↓ · Watch the hardware desk tour

▶ Watch the AgentDeck product tour  ·  🌊 Project website  ·  Devices  ·  Live preview  ·  Design system

AgentDeck · 2.0 product tour

AgentDeck product tour: sessions, collaboration, usage, real answers, evaluation and devices

A 2 minute 24 second tour of the whole workflow: parallel agent sessions, per-session history and CI waiting, collaboration, usage windows, original choices and one-time approval, voice and native agent controls, task evaluation and recorded evidence, then the roles of keypads, tablets, e-ink and small displays.

Actual Mac/iPad UI and production renderers use fictional activity; the phone control chapter uses a real local OpenCode execution. Static captures and older hardware photographs are labeled in the video. It was recorded on 2.0 builds, with English, Korean and Japanese captions; released downloads retain their channel versions. CloudKit/Watch remains a separate development experiment.

Watch the captioned product tour · Choice and approval detail · Aquarium detail · Device captures · Media provenance

Historical photographs and earlier recordings remain available. The device gallery labels its AI-assisted photo composite separately from actual app and firmware captures.

Start here

No extra hardware required. Choose a native dashboard or the terminal path.

Mac, iPhone, and iPad

Install from the App Store. The Mac app includes its own daemon and needs no Node.js. Enable your agent integrations in Settings, then run your agents normally. iPhone/iPad pair with an AgentDeck daemon on your network.

Requires macOS 26+ or iOS/iPadOS 17+. See the Apple guide.

Terminal, Windows, Linux, and external integrations

Requires Node.js 22, 24, or 26 and a supported agent on macOS 15+, Windows 11, or Linux.

npx @agentdeck/setup
agentdeck daemon install   # refresh hooks and start observation
claude                     # or: codex · opencode · kiro-cli

In another terminal:

agentdeck dashboard        # live sessions, aquarium, usage, and timeline

The Mac app can also attach to the CLI daemon for additional integrations, including Claude subscription gauges and ADB device support. Compare capabilities. For setup problems, run agentdeck diag agents or see installation and troubleshooting.

Existing managed sessions and remote attach

agentdeck claude, agentdeck codex, agentdeck opencode, and agentdeck monitor are a long-term compatibility path; no removal is planned. Ordinary local sessions should use the daemon with normal agent commands. Daemon-first equivalents exist for launch arguments (agentdeck run), deck ordering (agentdeck order) and remote sessions (agentdeck relay, experimental). Immediate terminal steering stays managed-only. See the CLI reference, remote attach guide, and decision record.

Choose your view

Start with your Mac or terminal; add screens and controls as needed. Mobile apps and hardware connect to a daemon on your computer.

Surface What it offers Setup
Mac · iPhone · iPad Native dashboard and 3D aquarium; Mac includes the daemon Apple guide
Terminal Sessions, braille aquarium, usage and timeline TUI guide
Android tablet or e-ink reader A dedicated companion dashboard Android guide
Stream Deck · Ulanzi Physical session keys and supported controls Stream Deck · Ulanzi
ESP32 · TRMNL Compact panels, rotary controls and e-ink status boards Firmware guide
Pixoo · TC001 · Timebox · iDotMatrix Agent status on pixel displays Device guide
Vobot Mini Dock Live session and Claude and Codex usage display on a trusted LAN. The development build adds full-screen input alerts and a pulsing rear LED Vobot guide · Device photo

Controls require a supported session control path; other surfaces show status. D200X support covers LCD keys, not encoders. See the hardware compatibility matrix for model-specific limits.

Browse device photos and live previews →

A dashboard that feels alive

On your Mac, no extra hardware required. Watch the native app in action below.

Native 3D aquarium in motion — agent creatures and schooling fish; click to watch the current recording

Choose the familiar dashboard or the native 3D aquarium on Mac, iPhone/iPad, and Android LCD devices. Each session becomes a creature; its activity changes its movement, and nearby fish react. On mobile, tap empty water to hide the panels and move closer. Tap again to return. Existing preferences are preserved.

The preview above shows fictional activity in the real native app; no private workspace data is shown. Watch the current aquarium recording or take the hardware desk tour.

What it does

  • Follow parallel work. Session names, current tools, and a shared timeline explain what each agent is doing. Collaboration shows reported subagents, task relationships, and recent task history.
  • Notice when you are needed. Permission and input requests stand out. Answer or interrupt from supported surfaces when the session provides a real control path; otherwise prompts are display-only.
  • Keep usage visible. Claude, Codex, and z.ai/GLM gauges appear where the connected provider supplies usage data. Agent identity and model provider remain separate.
  • Choose your display. The 3D scene keeps up to eight foreground residents readable while the roster retains every session. E-ink uses a static rendered backdrop and restrained motion; compact panels keep their focused layouts.

Agents

AgentDeck is growing beyond coding workflows: Hermes Agent is supported as an opt-in preview (see the table). The table below lists current integrations and how each is observed.

Agent Observation
Claude Code Lifecycle hooks; primary supported integration
Codex CLI Lifecycle hooks and rollout logs
Codex Desktop Observed on macOS; Windows verification pending
OpenCode Observer plugin and native events
Kiro CLI / IDE Transcript observation; delayed, idle-only activity
Antigravity Passive observation through the CLI daemon
OpenClaw Experimental Gateway integration
Hermes Agent Preview: opt-in observer plugin (agentdeck hermes-observer, then hermes plugins enable agentdeck-observer); read-only

Run agents normally; AgentDeck reads their native events rather than scraping terminal screens. Kiro has no managed launcher, and the sandboxed Mac app needs a one-time folder grant for its transcripts. Detailed setup and limitations: configuration, Apple guide, and Gateway guide.

Releases

Compatibility and delivery channels

Use the channel for your device. Components with the same major version work together; target-specific updates can ship at different times. Delivery records distinguish published downloads, Store listings and pending review, with the date each was checked.

Product Install / update Release tag
Mac · iPhone · iPad App Store apple-v*
Android tablets and e-ink Google Play · signed GitHub APK android-v*
CLI + daemon npx @agentdeck/setup npm-v*
Stream Deck / Mini / XL / Plus / + XL Elgato Marketplace streamdeck-v*
Ulanzi D200H / D200X LCD keys Ulanzi Marketplace; D200X encoders are not supported ulanzi-v*
Vobot Mini Dock Wi-Fi pairing on stock firmware 1.2.6 or later Gallery 0.3.0 approved
ESP32 panels and TRMNL 7.5" Browser flasher · firmware releases esp32-v*

Update Android through your existing signing channel: Play and GitHub APK signatures can differ. Supported ESP32 boards offer Wi-Fi OTA after the first USB flash. See the changelog, all releases, and version policy.

Compatible Companion Projects

Independent integrations have their own releases and support. Both projects below are Community integrations; listing does not imply verified compatibility.

See the Surface Protocol for capabilities, runtime status, integration manifests, and conformance levels.

Documentation

Start with the website — puritysb.github.io/AgentDeck carries the rendered device catalog, live renderer previews, the design system, and build health.

Using it CLI reference · Configuration · Troubleshooting · Windows · Linux
Surfaces Hardware matrix · Stream Deck layout · Devices · ESP32 · Android · Apple · TUI
Internals Architecture · Surface protocol · Internal bridge protocol · Daemon · Gateway protocol · Testing
Evaluation Why APME · APME · Pipeline
Design DESIGN.md · Tokens · Resource map
Project Roadmap · Releasing · Changelog · Agent harness · AI-assisted maintenance

Automated security checks

Security scan CodeQL analysis Snyk monthly comparison Security pipeline health

Dependency audits and pairing-token detection run on PRs, master pushes and daily. CodeQL adds source analysis; Snyk provides a monthly comparison, and an hourly watchdog checks scan freshness. PR dependency review flags newly introduced high/critical advisories.

Badges show workflow status, not a vulnerability-free certification. CodeQL can complete with alerts; Snyk can complete with findings recorded for review. The report preserves each run's commit, date, scope and limitations.

Results and coverage · Security scanning runbook · Report a vulnerability privately

Development

pnpm install && pnpm build
pnpm -r --parallel dev
pnpm test

Build from source · Testing · Build health. Coding agents should start at AGENTS.md.

Community

Bug reports, hardware verification, documentation fixes, and focused pull requests are welcome. Read CONTRIBUTING.md, use SECURITY.md for private vulnerability reports, and follow the Code of Conduct. See AI-assisted maintenance for how maintainers review agent-assisted work.

License & attribution

MIT — see LICENSE.

Independent project. Not affiliated with Anthropic, OpenAI, Google, Elgato, DIVOOM, or any other third party referenced here. All trademarks belong to their respective owners. Full notices in ATTRIBUTION.md.

Explorer des projets similaires