Kiro CLI is an AI coding agent that lives in your terminal — it reads your codebase, writes code, runs commands, and asks before doing anything destructive. This guide gets you from zero to productive in about five minutes.
If you haven't installed yet, run the install command from the installation page and complete browser authentication. Then come back here.
Kiro CLI has a built-in guide agent that knows everything about the CLI — commands, tools, settings, workflows, and best practices. Instead of reading docs, you can just ask it.
After launching a session, type:
> /guide
Then have a conversation:
[guide] > What can you help me with? [guide] > How should I set up this project to work well with Kiro? [guide] > What's the best way to do code reviews with Kiro? [guide] > How do I connect my GitHub account?
The Guide agent draws from indexed documentation that matches your installed version. It can also create configuration files for you — agents, prompts, and steering files — right from the conversation.
Press Shift+Tab to return to your main agent when you're ready to start working.
Navigate to any project directory and launch Kiro:
cd my-project kiro-cli
The terminal UI opens — a rich chat interface with syntax highlighting, interactive panels, and visual tool progress. You're now talking to an AI agent that can read, write, and run commands in your project.
Type naturally. Kiro understands your project context and can take action:
> Explain the architecture of this project > Find all TODO comments and summarize them > Write a unit test for the auth module > Refactor the database connection to use connection pooling
Kiro reads your files, runs commands, and makes changes — asking for permission before anything destructive.
Point Kiro at specific files using @ with tab completion:
> Review @src/api/routes.ts for security issues
Press Shift+Tab to switch to Plan mode. Kiro outlines what it intends to do before making any changes — useful when you want to review the approach first.
[plan] > Migrate the Express routes to use async/await error handling
The agent produces a structured plan with steps. When you're satisfied, it switches to execution mode and carries out the work.
Prefix with ! to run commands without going through the agent:
!npm run build
When Kiro wants to run a command or write a file, you'll see an approval prompt. Press y to approve, n to reject, or Tab to drill into options and provide feedback.
Out of the box, Kiro reads your files and infers context. But you can make it significantly smarter by telling it about your project's conventions. Create a steering file:
mkdir -p .kiro/steering
Add .kiro/steering/project.md:
# Project Context - This is a TypeScript project using Next.js 15 - Use pnpm for package management - Follow the existing code style (Prettier + ESLint) - Write tests with Vitest - Prefer functional components with hooks
Kiro reads steering files automatically and follows your guidelines in every response. The difference is immediate — responses match your stack, your conventions, and your preferences.
See Steering for the full guide on writing effective steering files.
Slash commands control your session. Type / and the TUI shows autocomplete suggestions. Here are the ones you'll use most:
| Command | What it does |
|---|---|
/help | Opens a searchable panel of all commands |
/guide | Switches to the Guide agent for help with Kiro itself |
/context | Shows what files are in context and token usage |
/model | Switch between AI models |
/agent | Switch between agents |
/chat new | Start a fresh conversation |
Kiro ships with built-in agents for different tasks, and you can create your own:
Shift+Tab)/guide)Create custom agents for recurring workflows — code review, test writing, deployment, or anything specific to your team:
> /guide Create an agent for code review
The Guide agent generates a .kiro/agents/code-review.yaml file. Switch to it anytime with /agent swap code-review.
For larger tasks, Kiro can spawn subagents that work in parallel with isolated context. Press Ctrl+G to open the crew monitor and watch them work in real time.
See Subagents and Custom agents for the full guide.
Beyond the per-action approval flow, you have fine-grained control over what Kiro can do:
npm run commands) so you're not interrupted for routine operations/tools trust <name> to pre-approve specific tools, or /tools untrust <name> to require explicit permission/tools reset clears all runtime permissionsSee Permissions for the full trust model.
Kiro supports the Model Context Protocol (MCP), which lets it interact with databases, APIs, GitHub, and other services directly. Once you configure an MCP server, you can ask things like "list my open pull requests" or "query the staging database."
See MCP integration to connect your first server.
Shift+Enter to type multi-line promptsShift+Tab to switch to Plan mode (agent plans before acting)Esc to cancel the agent or close any panel/help to see all available commandsThe full keyboard reference is in Terminal UI.
Now that you're up and running, here's where to go deeper:
/guide
Quick start