Skip to content

Home

Jump to bottom
Liu edited this page Mar 18, 2026 · 8 revisions

Racc

A desktop control plane for orchestrating AI coding agents. Not an editor. Not an IDE. A session orchestrator.

What is Racc?

Racc is a standalone desktop application (Tauri) for individual developers who use terminal-based AI coding agents. Currently supports Claude Code, with Codex support planned. It provides visual management for multiple concurrent agent sessions — something the terminal alone cannot offer.

Three design principles — the "Three Nots":

  1. Don't rebuild a code editor — users already have one they love
  2. Don't lock into a specific agent — agent-agnostic PTY communication
  3. Don't replace existing tools — integrate with git, docker, native OS primitives instead

Quick Navigation

Page Description
Product Vision Core positioning, target users, design principles
Feature Specification P0 (MVP), P1, and P2 features in detail
UI Design Layout, panels, and interaction patterns
Cognitive Design Research Neuroscience and human factors research informing UI decisions
Technical Architecture System architecture, tech stack, and tradeoffs
Session Lifecycle State machine, creation flow, reconciliation
WebSocket Remote API External client integration via WebSocket
Headless Server Design Three-crate workspace, racc-server, and RaccTransport
RTK Token Optimization Automatic 60-90% token cost reduction for Claude Code sessions
Roadmap MVP scope and versioned milestones

Key Technical Bets

  • Tauri 2.x — Rust backend + React 19 frontend + xterm.js terminals, single-process architecture
  • Native PTY — tauri-plugin-pty for real-time terminal I/O (replaced tmux)
  • Agent-agnostic communication — Agents interact via standard PTY read/write (currently Claude Code, Codex planned)
  • Git worktrees — Code isolation per session, zero overhead
  • Zustand — Lightweight state management for frontend
  • SQLite — Session and repo persistence at ~/.racc/racc.db

One-Line Summary

Racc is "the next step for terminal agent users" — keep the full power of their favorite agents, add the orchestration and visibility they've always lacked.

Clone this wiki locally