StarAgent

StarAgent

简体中文 ⚠️ This project is currently intended for personal use and is under active development. A stable version will be released later. ⚠️ This project was primarily built with vibe coding and may contain potential bugs. Please keep this in mind before using it.

StarAgent is a local-first Agent Harness Launcher and an optional multi-Node Hub for managing coding agent sessions. It reflects my own best practices for using multiple agents: We need a lightweight persistent terminal wrapper for Codex / Claude Code that supports cross-machine connections and can be accessed from any device. Design Principles

It is built around a few practical needs that show up when using coding agents day to day: We often run multiple agent CLI instances in parallel across different working directories, each handling a separate task. We need one place to check their status and interact with them in real time. We want to interact with our agents anywhere, anytime, and on any device. And the sessions should be consistent. Agent CLI sessions should be long-lived, so we do not need to keep typing /resume.

Based on hands-on experience, StarAgent uses the simplest effective stack for this workflow, making it feel like managing a small team of coding agents: Platform-native persistent terminals. Windows Desktop hosts sessions with ConPTY; Linux, macOS, and server Nodes use long-lived tmux sessions. See SESSIONS.md for the session model. Cross-machine connectivity via Tailscale. Tailscale provides a secure and unified network layer across machines. See tailscale/README.md for the Tailscale setup. Unified management through a web dashboard. The web dashboard lets you control agents from any device with a browser, including phones and laptops, without installing extra software. Node-local account and launch control. Sign in or switch credential sources from the Harness page, browse the models exposed by each CLI, and choose model plus reasoning-effort defaults for new Sessions without rewriting vendor configuration files.

Running staragent opens the single-Node StarAgent Launcher, scoped directly to the current machine. When cross-machine management is needed, staragent hub keeps the existing Hub experience: choose a Node, then enter the same Node workspace used by Launcher. For the technical architecture, see ARCHITECTURE.md. Preview

The screenshots below use sanitized demo data. The anime-inspired background is optional — upload any image from the theme menu, then pair it with a color theme and glass surfaces.

Select a connected Node, then manage that machine's sessions in a focused workspace:

Each session includes a lightweight chat console for interacting with agents, plus a Terminal and File Explorer. Session detail pages keep an IM-style switcher on the left for moving between that Node's sessions without returning to the session table; on narrow screens it collapses into a drawer.

Inspect Agent CLI availability, login state, installed Skills, streamlined install/update paths, and launch presets on the selected Node: