The terminal coding-agent space filled up fast — Claude Code, Codex, Gemini CLI, OpenCode — so the real question for any newcomer is what it does differently. The answer here is to treat the shell as a first-class surface rather than a sandbox: a single Ctrl-X toggle drops you into raw shell commands and back without leaving the agent loop, and the whole thing speaks the Agent Client Protocol so it can live inside Zed or JetBrains instead of only the terminal.
What Sets It Apart
- Shell-native loop: Ctrl-X lets you mix agent prompts and hand-typed commands in one session instead of context-switching to a separate terminal — handy when you trust yourself for some steps and the agent for others.
- Runs where you work: it ships as a CLI, a VS Code extension, and an ACP server for Zed/JetBrains, so the same agent backend is reachable from three entry points.
- First-party Moonshot tooling: built and maintained by the team behind the Kimi models, with MCP support (HTTP + stdio transports, OAuth) for wiring in external tools, plus a zsh plugin for shell integration.
Who It's For
Great fit if you already use Kimi models, live in the terminal, and want an agent that blends manual shell work with autonomous multi-step execution — or if you want one agent reachable from both your editor and your shell. Look elsewhere if you need a mature, heavily documented ecosystem: it's young (still mid-rename from Kimi CLI to Kimi Code CLI), Python-first, and moving fast through frequent releases, so expect rough edges over rock-solid stability.
