AI Desk Card Agent Hardware Skill

AI Desk Card Agent Hardware Skill

AI Desk Card is an open-source Agent Skill from op7418 that turns an M5Paper V1.1 4.7-inch e-ink panel into an ambient desk display. The key idea is not another fixed IoT dashboard: the agent decides which widget matters now, renders it through a local daemon, and pushes it to the physical screen.

Ce que fait ce skill

• Install the ai-desk-card Skill for Claude Code, Codex, Cursor, Gemini CLI, Aider, or a custom shell-capable agent.
• Guide first-time setup: detect PlatformIO, USB connection, daemon state, Wi-Fi state, and hardware readiness.
• Build and flash firmware, flash CJK fonts, start the local daemon, provision Wi-Fi, and push the first widget.
• Use 16 widget types such as weather, calendar, todo, focus, inbox, PR queue, git status, AI status, break reminder, next meeting, deadlines, scratch notes, and business-card sleep mode.
• Let the agent write ~/.ai-desk-card/interests.yaml, register a loop or cron schedule, and adapt the screen based on work context.

Cas les plus adaptes

• Makers who already use Claude Code, Codex, Cursor, or another agent and want a physical ambient output.
• Developers maintaining GitHub projects who want PR, issue, CI, and focus-state signals without opening GitHub every few minutes.
• Writers, operators, and founders who want calendar, todo, weather, deadlines, and reminders in peripheral vision.
• People who like e-ink because it is quiet, non-glowing, and good for glanceable information rather than constant interaction.

Flux dutilisation

1. Install the skill with npx skills add https://github.com/op7418/ai-desk-card --skill ai-desk-card or clone the repository into the local skills folder.
2. Ask the agent to set up the card; it probes current state before acting.
3. Plug in M5Paper V1.1 with a USB-C data cable. The agent handles PlatformIO, build, firmware upload, CJK font upload, daemon startup, and Wi-Fi provisioning.
4. Choose widgets and cadence. The skill writes local preferences and can schedule refreshes through the agent loop, cron, or no-AI fallback scripts.
5. Use natural language for daily operations: show weather, pin today’s schedule, display PR queue, sleep with business card, or diagnose connection issues.

Cas dusage pratiques

• Calendar and todo card: the screen shows meetings and tasks that change as the agent updates calendar or todo sources.
• Quiet-hours business card: e-ink retains the final frame with zero power, so a QR business card can stay visible while the device sleeps.
• GitHub ambient queue: PRs, mentions, issues, CI failures, and critical labels can be pushed without turning GitHub into a constant distraction.
• Health and context nudges: break reminders, weather, deadlines, and low-priority but useful signals stay visible without popup fatigue.
• Hardware-as-agent-output: the M5Paper remains simple while the agent, memory, CLI tools, and APIs provide the intelligence.

Limites et risques

• It requires hardware: M5Paper V1.1 is the primary target; M5Paper V1.0 likely needs tuning, and M5Paper S3 / Inkplate / Waveshare are roadmap or porting work.
• E-ink is slow and monochrome; it is not suitable for video, animation, realtime stock ticks, or rich touch interfaces.
• First setup still touches firmware, USB, Wi-Fi, Python, PlatformIO, and local daemon permissions, so the agent needs shell access and human approval for risky commands.
• ESP32 Wi-Fi is 2.4 GHz; 5 GHz-only networks will not work.
• The repository is very new, so teams should check license, security posture, hardware compatibility, and maintenance before treating it as production infrastructure.

Sources consultees

• GitHub repository: https://github.com/op7418/ai-desk-card
• Official README: https://github.com/op7418/ai-desk-card/blob/main/README.en.md
• Chinese README: https://github.com/op7418/ai-desk-card/blob/main/README.md
• X article by op7418 / 歸藏: https://x.com/op7418/status/2057776589027594406
• M5Paper documentation: https://docs.m5stack.com/en/core/m5paper

FAQ

Does the agent really install and flash the device? The official README describes an agent-led flow: probe state, install missing tooling when approved, compile firmware, flash fonts, start the daemon, ask for Wi-Fi, and push the first widget.

Is it cloud-based? The project is designed around a local daemon on 127.0.0.1:9877 and LAN device communication. Verify your own integrations because calendar, weather, GitHub, Feishu, or other data sources may involve their own network calls.

What makes it different from a fixed dashboard? Traditional dashboards ask users to configure widgets. AI Desk Card treats the visible screen as a scheduling decision: the agent can decide what matters based on memory, calendar, repo status, and current work mode.