A native terminal for running parallel AI coding agents. Each agent gets an isolated git worktree in a split pane with a live status indicator, sessions keep running after you close the app, and you can steer agents from your phone. Claude Code, Codex and any CLI agent. Rust on Tauri v2 with libghostty-vt and a custom wgpu render pipeline, no Electron. Native on macOS, Windows and Linux, plus a headless Rust PTY daemon for remote Linux boxes. Offscreen render benchmark: p50 3.10 ms, p95 4.34 ms at 800x480 on an M4 Max.

AgentsDevTools

RustTerminal

Looking forBeta users Feedback

About

I build Ferryx (ferryx.dev), a native terminal for running AI coding agents. It's not another chat window: it's the workspace layer under Claude Code, Codex and the rest.

The problem: coding agents are long-running, parallel and interrupt-driven. Five agents running means five terminals, five worktrees and no idea which one is waiting on you. Ferryx gives each agent an isolated git worktree in a split pane with a live status indicator, keeps sessions alive after you close the app, and lets you steer them from your phone.

It's Rust all the way down: Tauri v2 shell, libghostty-vt as the terminal parser, and my own wgpu render pipeline instead of Electron. Offscreen render benchmark is p50 3.10 ms / p95 4.34 ms for 50 frames at 800x480 on an M4 Max. Native builds on macOS, Windows and Linux, plus a headless Rust PTY daemon and CLI for remote Linux boxes so you can attach without SSH gymnastics.

I'm in Seoul, in it solo. What I want from this room: anyone else running multiple agents in parallel who has an opinion on worktree isolation and agent supervision, and anyone who has shipped a native desktop app and can tell me where Tauri stops being the right call.