Multi-task split chat, integrated terminal, file tree and Git review in one window. The command-line agent experience, brought into a real IDE.
Each task is an independent, resumable CLI session. The app never calls the API directly — all model power comes from your local Claude Code / Codex.
2D tiling main area, cross-workspace mixing, draggable splitters, real-time token streaming.
A real xterm.js + PTY terminal alongside chat and files.
Browse code and diffs; @-mention to feed files to the AI.
Branch / worktree switching, change review panel, ✨ one-click commit messages.
Visual management of subagents and model providers; extend the AI on demand.
Bind a bot to create and drive tasks from your phone, and read AI replies.
Each task/draft picks its own CLI family, model, reasoning effort and permission mode. Streaming, interactive approval, graceful interrupt, exact history restore, subagent cards and forking — fully aligned across both.
One process per task · lazy start · idle reapFree · cross-platform · auto-update
⚡ Fast download from a China-based server · not via GitHub
LoongCode ships no model of its own; it relies on your local Claude Code CLI (or Codex). Make sure the CLI is installed, on PATH and logged in — or one-click install it under Settings → Dependencies.
GitHub Releases are often slow or fail in mainland China. Installers here are hosted on a China-based server and synced automatically from the official release repo.
Configure an http / https proxy in app settings (the spawned CLI does not support SOCKS).