Kimi Code Cheat Sheet

Start

kimi                  # start interactive TUI
kimi -c               # continue most recent session in this directory
kimi -S [id]          # resume a session by ID, or pick interactively

Session

/new          # fresh session, discards current context (alias: /clear)
/sessions     # browse/restore historical sessions (alias: /resume)
/fork         # fork current session preserving full history
/undo [n]     # undo recent prompts (not code changes)
/exit         # exit (aliases: /quit, /q)

Context / Compact

/compact [instruction]   # compact context, optional hint of what to preserve
/init                    # analyze codebase and generate AGENTS.md
/status                  # version, model, cwd, permission mode
/usage                   # token usage, context consumption, quota

AGENTS.md at the project root is injected as project instructions; global instructions live in ~/.kimi-code/AGENTS.md and ~/.agents/AGENTS.md.

Plan

/plan [on|off]    # toggle Plan mode
/plan clear       # clear current plan

Or press Shift-Tab to toggle Plan mode. Start in Plan mode with kimi --plan.

Goal

/goal <objective>    # persistent objective across auto-continuing turns
/goal status         # show goal, elapsed time, turns, tokens
/goal pause          # pause the current goal
/goal cancel         # cancel the current goal

Review

No direct equivalent verified.

Permissions

Three modes: Always Ask (manual, default; every edit/command asks), Ask When Needed (yolo; routine ops auto, risky still asks), Never Ask (auto; fully unattended).

/permission    # select permission mode
/yolo          # switch to Ask When Needed (alias: /yes)
/auto          # switch to Never Ask

kimi --yolo    # start in Ask When Needed mode
kimi --auto    # start in Never Ask mode

Warning: --yolo skips approval for regular tool calls including file writes and shell commands — use only in trusted working directories. --auto is fully unattended: the agent never asks and dangerous commands run without interruption.

Configuration

~/.kimi-code/config.toml    # main config (KIMI_CODE_HOME relocates the data root)
~/.kimi-code/tui.toml       # TUI preferences

/settings    # settings panel in the TUI (alias: /config)
/reload      # apply latest config.toml and tui.toml without restart

kimi doctor    # validate config.toml / tui.toml without the TUI

MCP

MCP servers are configured in mcp.json, not config.toml: ~/.kimi-code/mcp.json (user) and .kimi-code/mcp.json (project; project wins on name conflicts).

/mcp           # list MCP servers and connection status
/mcp-config    # add/edit/delete servers, MCP OAuth login

{
  "mcpServers": {
    "filesystem": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-filesystem", "/tmp"] },
    "linear": { "url": "https://mcp.linear.app/mcp" }
  }
}

Non-interactive

kimi -p "summarize this repo"                          # one-shot, streams to stdout
kimi -p "query" --output-format stream-json            # JSONL output

Runs with auto permission (no approval prompts); static deny rules still apply. Goal-mode exit codes: 0 complete, 3 blocked, 6 paused.

---

Last verified: 2026-09-07

AI coding tools evolve quickly. If you find outdated information, please open an issue or pull request.