Kimi CLI cheat sheet

Every command, flag and slash command for kimi — Moonshot's terminal agent — doubles as a shell (Ctrl-X), MCP + ACP, powered by Kimi K2. 18+ commands, verified July 2026.

Rebranding to Kimi Code (github.com/MoonshotAI/kimi-code) — the kimi command + kimi-cli package still work and installing Kimi Code auto-migrates config/sessions.

Install

curl -LsSf https://code.kimi.com/install.sh | bash     # macOS / Linux (installs uv + kimi-cli)
uv tool install --python 3.13 kimi-cli                 # if you already have uv (Python 3.12–3.14)
kimi --version   ·   uv tool upgrade kimi-cli

Start & run

kimi                         # interactive session in the current dir
kimi login                   # OAuth login (or /login in-session)
kimi --print -p "query"      # headless: print and exit (implies auto-approve)
echo "explain this" | kimi --print
kimi --continue              # continue the last session here
kimi --resume <id>           # resume a specific session (no id → picker)

Key flags

Flag	What it does
`-p, --prompt`	Run one prompt then exit (`-c, --command` alias)
`--print`	Non-interactive print mode (implies `--afk` auto-approve)
`--quiet`	= `--print --output-format text --final-message-only`
`-C, --continue`	Continue the previous session in this dir
`-S, --session [id]` / `-r, --resume [id]`	Resume a session (or open the picker)
`-m, --model <name>`	Override the default model
`--thinking` / `--no-thinking`	Toggle thinking mode
`--plan`	Start in plan mode (read-only exploration)
`-y, --yolo`	Auto-approve all tool calls

In-session slash commands

Command	What it does
`/login` (`/setup`)	Log in / configure platform + model
`/model`	Switch model and thinking mode
`/init`	Analyze the project → generate an `AGENTS.md`
`/plan [on\|off\|view]`	Toggle plan mode
`/new` · `/sessions` (`/resume`)	New session · list/switch
`/compact [instructions]`	Compact context to save tokens
`/undo` · `/fork`	Roll back a turn · fork the session
`/mcp` · `/hooks`	Connected MCP servers · hooks
`/skill:<name>` · `/flow:<name>`	Load a skill · run a flow

Keys: Ctrl-X toggles agent ↔ shell mode · Shift-Tab plan mode · Ctrl-S steer the running turn.

Config

~/.kimi/config.toml          # providers, models, default_model (TOML or JSON)
~/.kimi/                     # sessions, credentials, logs, mcp.json

export KIMI_API_KEY="…"      # KIMI_BASE_URL, KIMI_MODEL_NAME also honored
# base URL: https://api.moonshot.ai/v1 (global)  or  …moonshot.cn/v1

Common one-liners

kimi --print -p "Add type annotations to src/utils.py"
git diff --staged | kimi --quiet -p "Write a Conventional Commits message"
kimi mcp add -t http context7 https://mcp.context7.com/mcp -H "CONTEXT7_API_KEY: …"
kimi --resume                # pick a past session