AI Maestro: A Safe Getting-Started Guide to the Multi-Agent Orchestration Dashboard

Who This Is For

• Developers who run multiple AI coding agents at the same time (such as Claude Code, Codex, Grok Build, etc.) and want to manage all of these agent sessions from a single web dashboard instead of switching between multiple terminals.
• Users who want their agents to have extended capabilities such as memory search, code graph queries, and agent-to-agent messaging.
• Prerequisite: installation is done on your local machine with a single shell command. Because the app has no application-layer authentication, it must only be used on a trusted network (such as a Tailscale private network), and port 23000 must not be exposed to untrusted networks. If you cannot guarantee the network boundary, or you don't accept this security model, this tool is not for you.
• Official evidence does not state the minimum supported operating systems or agent version requirements; this information is currently unknown.

First Try in 3 Minutes

The recommended main path is the remote shell installation given in the official documentation:

curl -fsSL aimaestro.sh | bash

1. Prerequisite: a machine that can run a shell, and it must be within a trusted network you control.
2. Install: run the command above in a terminal. It runs the official remote-install script. Note: this downloads and executes a script from the network, so confirm you trust the source before installing.
3. After installation: once started, it provides a web-based terminal interface and manages each AI agent session through tmux (xterm.js terminal + node-pty + WebSocket streaming output).
4. For your first try, pick a low-risk task: open an existing agent session in the dashboard and have it do a read-only analysis, for example summarizing the structure of a code file. Which tasks a specific agent supports depends on that agent itself; official evidence does not list them.
5. Success signal: you can see the agent's terminal session streaming output in real time in your browser, and the agent produces the analysis result. If the dashboard can display multiple agent sessions at once, orchestration is working.

The official documentation also mentions managing multiple agents from one dashboard, but no other installation methods are described, so this guide keeps only the single official path above.

Use Cases

Case 1: View multiple agent sessions in one place. Scenario: You are running Claude Code and another agent at the same time and want to see both in one interface. How: After installing and starting the app, open the dashboard in your browser and connect the agent sessions you want to manage (officially, it manages sessions through tmux). Expected result: you can see real-time output from each session in a single web interface via xterm.js terminals.

Case 2: Have an agent do a read-only code analysis. Scenario: You want to verify the orchestration environment works without letting the agent modify any files. How: Send an instruction to an agent in the dashboard, for example "analyze the structure of a file under the src directory and summarize it." Capabilities used: the dashboard's terminal session itself; whether memory search or code graph queries are invoked depends on the skills loaded into that agent — official evidence does not state what is loaded by default. Expected result: the agent outputs the analysis text in the terminal; as long as there are no file-modifying operations, this is a reversible, observational task.

Limits of Use

• Network: the app binds to 0.0.0.0 by default and has no application-layer authentication. The security model is "the network is the trust boundary"; the official documentation requires using it together with a network boundary such as Tailscale, and port 23000 must not be exposed to untrusted networks.
• Side effects and confirmation points: installation executes a remote script; agent sessions may run commands or access files, and their exact behavior depends on the agents you connect. Official evidence does not state any additional restrictions this app places on agent operations, and unknown information must not be treated as "no risk."
• Runtime environment: requires tmux and a browser with WebSocket support; official evidence does not list operating system requirements.
• Explicit official limitation: the lack of application-level authentication is a design decision, not evidence of a configurable option.
• Not yet confirmed: the supported text/agent language range, the complete list of supported clients, and upgrade/uninstall procedures — none of these are covered by official evidence.