Open-Source USB Desktop Screen: Real-Time Monitoring of Multiple AI Coding Agents
managents uses a small ESP32 screen over USB to display the status of multiple AI coding agents in real time, processing data locally and open-sourced under MIT.
managents is an MIT-licensed open-source project that uses a 4-inch ESP32 screen (LCDWiki E32R40T) connected to a computer via USB to display the status of multiple AI coding agent sessions in real time, freeing developers from repeatedly switching between terminals. A lightweight Go helper runs on the computer, automatically detecting Claude Code and OpenCode sessions and sending only metadata such as running, waiting for approval, completed, or failed—without reading code content, with all data processed locally. The project offers a one-line install command and `managents flash` firmware flashing (about 13 seconds), plus OpenSCAD 45-degree stand files. The repository's `AGENTS.md` supports bootstrapped installation by AI agents. The author is seeking community help to port to Windows and extend support to Codex CLI, Gemini CLI, Aider, Cursor, Copilot, and more.
Freed from Terminal Switching
When running multiple AI coding agents, developers often need to switch back and forth between terminals to confirm which task has finished, which is waiting for approval, and which has failed. One developer shared a homemade hardware solution on Reddit: a 4-inch ESP32 screen (LCDWiki E32R40T) connected to a computer via USB, displaying the status of all AI agent sessions in real time. The project is called managents, open-sourced under the MIT license, with both code and hardware design files published on GitHub.
The core logic is: a lightweight Go helper program runs on the computer, responsible for discovering local AI agent sessions and sending only status information over USB. Data stays entirely local and is never uploaded to the cloud.
Hardware and Software Architecture
Custom Firmware and 3D-Printed Enclosure
The screen firmware is custom-developed and runs on the ESP32. The author provides OpenSCAD files for 3D-printing a 45-degree stand for easy desktop viewing. Installation is simplified to a one-line command, followed by the managents flash command to flash the display program, which takes about 13 seconds.
Helper Program and Privacy Design
The Go helper currently supports macOS and Linux, and automatically detects Claude Code and OpenCode sessions. It only extracts status metadata (such as running, waiting for approval, completed, failed) and does not read or transmit code content. This design provides convenience while avoiding leakage of sensitive information.
How the Go Helper Detects Sessions
The managents Go helper can automatically discover Claude Code and OpenCode sessions because these AI coding agents expose locally observable state information during operation, such as process trees, local log files, Unix domain sockets, or structured session state files. The helper runs as a persistent background process, periodically scanning these sources, normalizing each session into a limited set of status enumerations (running, waiting for approval, completed, failed), and then sending them to the ESP32 screen via a USB serial protocol.
This "read-only metadata, no touching code content" design means the helper has a narrow permission boundary—even if maliciously invoked, it would be difficult to steal source code. The trade-off is a strong dependency on each agent's own observability interfaces: once an agent updates and changes the format or path of its state files, the detection logic may fail and require synchronous updates. This is also why the author seeks community help to extend support to new agents like Codex CLI and Gemini CLI—each agent's detection method needs to be adapted individually.
Bootstrapped Installation for AI Agents
The repository includes an AGENTS.md file, which means users' AI coding agents can handle instructions like "install managents and flash my display" on their own. Before execution, each command is shown to the user for confirmation. This bootstrapped installation lowers the configuration barrier and demonstrates the new role of AI agents in the development toolchain.
How AGENTS.md and Agent Bootstrapped Installation Work
AGENTS.md is a structured instruction file in the repository aimed at AI coding agents—essentially a README for agents. When a user issues a command like "install managents and flash my display" in tools such as Claude Code or Cursor, the agent reads the steps, dependencies, and commands defined in AGENTS.md, and executes them in order: cloning the repository, installing the Go helper, invoking esptool via managents flash to write firmware to the ESP32, and so on.
To control risk, managents adopts an interactive mode of "show each command first, then confirm." This prevents the agent from executing arbitrary system commands without the user's knowledge. This design extends the AI agent from a "coding assistant" to an "operator capable of configuring development environments," and incidentally demonstrates that in the AI agent toolchain, documentation formats themselves are evolving from human-readable to machine-executable.
Current Support and Community Collaboration Needs
The project currently supports Claude Code and OpenCode, but the author explicitly lists three areas where community assistance is needed.
Windows Support
The helper is only compatible with macOS and Linux. Windows users who want to test or port it can contact the author directly. Given Windows's share among developers, this gap may limit the project's reach.
Extending to More AI Agents
The author hopes to support more agents such as Codex CLI, Gemini CLI, Aider, Cursor, and Copilot. The contribution guide (CONTRIBUTING.md) explains the process for adding a new agent: tell the author which agent you use, and he will research the detection method. This open collaboration model helps the project quickly cover mainstream tools.
Differences Between Codex CLI, Gemini CLI, Aider, Cursor, and Copilot
Although these tools are all categorized as AI coding agents, their forms vary greatly: Codex CLI and Gemini CLI are pure terminal command-line tools, with session states typically existing as local processes or standard output streams; Aider uses Git commits as its unit of work, with state inferable from the repository and terminal logs; Cursor and GitHub Copilot are deeply embedded in IDEs, with states often exposed through the editor extension's local API or configuration directory.
Therefore, "extending support" is not simply adding a string—it requires individually researching what form each agent's state takes locally, then writing corresponding detection logic in the Go helper. This also explains the author's approach in the contribution guide: "tell the author which agent you use, and he will research the detection method." Detection logic is hard to generalize; only by first determining an agent's state output can it be integrated into managents' unified state model.
Feature and Feedback Solicitation
The author asks the community whether they want a similar product and what other information the screen should display—current task, token usage, elapsed time, etc. This indicates the project is still in an early stage, with its form shaped jointly by user needs.
Why It's Worth Paying Attention To
Multi-AI-agent parallel workflows are becoming the norm for some developers. When running three or four agents simultaneously, the cognitive burden shifts from "writing code" to "managing agents." managents uses a desktop secondary screen to visualize status, reducing context-switching costs.
The project also reflects a trend in the AI coding tool ecosystem: auxiliary tools around agents are beginning to emerge, from monitoring and orchestration to debugging. The combination of open-source hardware and AI agents offers individual developers the possibility of customizing their work environment.
Repository address: https://github.com/tonylook/managents. Stars, Issues, and PRs are welcome.
One-line Summary
managents uses a small ESP32 screen connected via USB to display the status of multiple AI coding agents in real time, processing data locally and open-sourced under MIT.
Paragraph Summary
managents is an MIT-licensed open-source project that uses a 4-inch ESP32 screen (LCDWiki E32R40T) connected to a computer via USB to display the status of multiple AI coding agent sessions in real time, freeing developers from the cost of repeatedly switching between terminals. A lightweight Go helper runs on the computer, automatically detecting Claude Code and OpenCode sessions and sending only metadata such as "running, waiting for approval, completed, failed" without reading code content, with all data processed locally. The project provides a one-line installation command and managents flash firmware flashing (about 13 seconds), along with OpenSCAD 45-degree stand files. The AGENTS.md in the repository supports bootstrapped installation by AI agents. The author is seeking community help to port to Windows and extend support to Codex CLI, Gemini CLI, Aider, Cursor, Copilot, and more.
Related articles
"Black October": Did AI Uncover 3% of All Critical Vulnerabilities Ever?
A Reddit post claims 3% of all critical software vulnerabilities ever were found in the past month, dubbed 'Black October.' We break down the statistical issues, how AI-assisted code auditing accelerates discovery, and how developers should respond.
Microsoft Project Gecko: Using AI to Bridge Language Gaps for African Farmers
Microsoft's Project Gecko launches PharmaChat, letting African farmers ask questions in their native language and get precise video clips and farming book chapters via AI. The project also releases three interactive manuals to advance inclusive AI.
Musk's Space Dominance: How Starlink Throttling Undermines Ukraine's Defense
SpaceX's Starshield throttling disabled Ukraine's high-speed interceptor drones, exposing how Musk's private control over low Earth orbit communications affects battlefield dynamics and national security, and highlighting the risks of commercial space militarization and space governance dilemmas.