Kiro Dev Stack
BeginnerAWS's spec-driven IDE, planning features before the agent codes, on Claude, GPT, or open-weight models.
Published 12 October 2026
About Kiro Dev Stack
Kiro is AWS's agentic IDE, built on VS Code's open-source core, and its agent is part of the editor rather than a plugin. What sets it apart is spec-driven development: for a feature, the agent first writes the requirements as user stories with acceptance criteria, then a technical design, then a task list, each saved as a file in the repository and approved before the next step. It then works through the tasks, running independent ones in parallel. Smaller changes skip the plan and go through chat, with every edit shown as a diff and checkpoints to roll back. Steering files hold the project's rules, hooks run the agent when files change, and powers load MCP servers and skills when a task needs them.
Models come from Kiro's own picker and are paid for in Kiro credits, with no separate API keys. Claude is the default here and the family Kiro leads with; OpenAI's GPT-5.6 models sit beside it for demanding work, and open-weight models from DeepSeek, Qwen, GLM, and MiniMax cost a fraction of the credits for routine tasks. An Auto setting picks a model per request. The free plan covers the open-weight models and one Claude Sonnet model, and the paid plans add the rest.
Kiro runs on your own computer by default, which has to stay on while the agent works there. For work that should continue with the laptop closed, paid plans add cloud sessions: the same agent runs in an isolated sandbox in Kiro's cloud against a GitHub or GitLab repository and opens a pull request when it's done. A session can start in the IDE, the Kiro CLI, or Kiro Web in a browser, and be picked up from any of them, or from an iOS app that's still in early access. Developers who prefer their own hardware can run the Kiro CLI headless on a server.
GitHub is the default for version control, with GitLab supported in cloud sessions too. Claude Code, Codex, or OpenCode can optionally run beside Kiro's agent, and an AI code reviewer, CI/CD, and Docker are common additions.
Key Features
- ✓Feature specs that write requirements, a design, and a task list as files in the repository before any code
- ✓Bugfix specs that diagnose a defect before the agent edits anything
- ✓Claude by default, with GPT-5.6 and open-weight DeepSeek, Qwen, GLM, and MiniMax models in one picker
- ✓Cloud sessions that keep working in a sandbox with the laptop closed and open a pull request
- ✓The same agent and .kiro configuration in the IDE, the Kiro CLI, and Kiro Web
- ✓Steering files, agent hooks, and powers that bundle MCP servers and skills
- ✓Open VSX extensions and imported VS Code settings, so an existing setup carries over
When to Use Kiro Dev Stack
- →Planning a multi-file feature as reviewable requirements and a design before the agent writes code
- →Giving reviewers the spec files alongside the pull request so they can check the code against the plan
- →Handing a task to a cloud session at the end of the day and reviewing the pull request in the morning
- →Stretching monthly credits by running routine edits on an open-weight model and hard ones on Claude
- →Diagnosing a production bug with a bugfix spec before any code changes
Pros
- Specs put the plan in writing, so a large change can be checked before code exists
- One subscription covers Claude, GPT, and open-weight models, with no API keys to manage
- Cloud sessions on GitHub or GitLab, started from the editor, the terminal, or a browser
- Low-multiplier open-weight models make routine work cheap within the plan's credits
- Built on VS Code's core, so settings and most extensions carry over
Cons
- Credit use per request varies with the model and the task, so monthly cost is hard to predict
- The free plan's 50 credits cover only a few real sessions, and cloud sessions need a paid plan
- Writing a spec costs time and credits that a small fix doesn't need
- Extensions come from Open VSX, which lacks some Microsoft ones such as the official Remote SSH and Pylance
- A separate agent like Claude Code or Codex needs its own subscription on top of Kiro's
LLM Options for Kiro Dev Stack
Kiro Dev Stack powered by Claude
Claude is the default here and the family Kiro leads with. Sonnet 5.5 costs 1.3 times the Auto rate in credits and suits most spec and chat work; Opus 5.5 at 2 times takes the hardest multi-file changes, and Haiku 4.5 at 0.4 times quick edits. The free plan includes Claude Sonnet 4.5.
Kiro Dev Stack powered by OpenAI
OpenAI's GPT-5.6 family sits in the same picker on paid plans, billed in Kiro credits: Sol at 4.4 times the Auto rate for the most demanding work, Terra at 2.2 times as a balanced pick, and Luna at 0.6 times for fast, cheap edits. Useful as a second opinion when Claude gets stuck on a task.
Kiro Dev Stack powered by DeepSeek
DeepSeek 3.2 runs at a quarter of the Auto rate in credits, on the free plan as well as the paid ones, which makes it a cheap choice for routine edits, tests, and refactors while a Claude or GPT model handles the planning and the harder spec tasks.
Kiro Dev Stack powered by Qwen
Qwen3 Coder Next is the cheapest model in Kiro's picker at 0.05 times the Auto rate, open-weight and tuned for coding. It stretches a small plan, or the free plan's 50 credits, furthest on simple, well-scoped tasks, though it trails Claude on long multi-file changes.
Kiro Dev Stack powered by GLM
GLM-5 from Z.ai runs at half the Auto rate in credits, an open-weight model with strong agentic coding results that sits between the budget models and Claude in price. Available on the free plan, and a reasonable default for teams watching credits closely.
Kiro Dev Stack powered by MiniMax
MiniMax M2.5 runs at a quarter of the Auto rate, and the older M2.1 at 0.15 times, both open-weight models built for agentic coding and tool use. A low-cost pick for long task lists in a spec, where many steps are routine and credits add up.
These are highlighted picks. To see all the tools, check the LLM category.
Version Control Options for Kiro Dev Stack
Kiro Dev Stack with GitHub
GitHub is the default here, where most repositories and AI reviewers already live. Kiro's cloud sessions clone a GitHub repository, work on a branch, and open a pull request, and they can start from an issue as well as from a prompt.
Kiro Dev Stack with GitLab
GitLab bundles source control, merge requests, and CI/CD in one platform you can self-host. Kiro's cloud sessions connect to GitLab projects with a personal access token and open merge requests, and one session can change repositories on GitHub and GitLab together.
Kiro Dev Stack Add-ons
Each addition below extends this stack with a capability the base stack works fine without. None are required: include the ones your product actually needs when building this stack, and skip the rest.
Coding Agent Add-ons
Add a separate coding agent when you want another vendor's agent working alongside the one built into the editor, either through its own extension or in the integrated terminal.
Kiro Dev Stack with Claude Code
Add Claude Code when you want Anthropic's agent on a task beside Kiro's: its official extension is published on Open VSX, so it installs in Kiro's sidebar, or the CLI runs in the integrated terminal. It needs its own Claude plan, from $20/mo, since Kiro's credits don't cover it.
Kiro Dev Stack with Codex
Add Codex when you want OpenAI's agent alongside Kiro's: the official Codex extension installs from Open VSX, and it runs on a ChatGPT plan you may already have, including the free tier. Kiro's spec files are plain Markdown, so Codex can read them as context too.
Kiro Dev Stack with OpenCode
Add OpenCode in Kiro's integrated terminal when you want an open-source agent that runs any provider's model with your own API key, or a local model, outside Kiro's credit system.
These are highlighted picks. To see all the tools, check the AI Coding Agents category.
Code Review Add-ons
Add code review when you want an AI reading every pull request before it merges: it comments on the diff so bugs, security issues, and inconsistencies surface before a human has to catch them.
Kiro Dev Stack with CodeRabbit
Add CodeRabbit when pull requests from Kiro's cloud sessions should get a reviewer outside the agent: it installs as a GitHub or GitLab app, summarizes the diff, and leaves line-level comments, with a free tier for open-source repos.
Kiro Dev Stack with Greptile
Add Greptile when review should read the whole repository rather than only the diff: it indexes the codebase so its GitHub and GitLab comments carry wider context, and a self-hosted deployment keeps the code on your own infrastructure.
Kiro Dev Stack with Cursor Bugbot
Add Cursor Bugbot when a bug-focused reviewer is enough: it reviews GitHub or GitLab pull requests whichever agent wrote them, though it bills through a Cursor plan rather than a subscription of its own.
CI/CD Add-ons
Add CI/CD when you want a dedicated pipeline for running tests, linting, or multi-stage builds before a deploy goes out. Many hosting platforms already redeploy automatically on every push on their own — a CI/CD tool adds the most value on top of that by gating the deploy on a passing test suite, and matters even more when the hosting choice does not auto-deploy at all, such as a self-hosted server.
Kiro Dev Stack with GitHub Actions
Add GitHub Actions when every push should run the tests, including the pull requests Kiro's cloud sessions open. The Kiro CLI's headless mode also fits a workflow step, authenticated with an API key.
Kiro Dev Stack with GitLab CI/CD
Add GitLab CI/CD when the project lives on GitLab: pipelines defined in a .gitlab-ci.yml file run the tests on every merge request, on GitLab's runners or your own.
These are highlighted picks. To see all the tools, check the CI/CD Pipelines category.
Containerization Add-ons
Add containerization when you want the app packaged the same way across local development, staging, and production, or need to deploy somewhere that isn't a managed serverless platform.
Kiro Dev Stack with Docker
Add Docker when the project needs the same environment on a laptop, in CI, and in production, or when a new teammate should start with one docker compose up. Kiro's agent can write the Dockerfile and Compose file as tasks in a spec.
These are highlighted picks. To see all the tools, check the Containerization category.
Frequently Asked Questions about Kiro Dev Stack
Can Kiro keep working when my computer is off?
Yes, through cloud sessions on a paid plan. The agent runs in an isolated sandbox in Kiro's cloud with the repository cloned from GitHub or GitLab, so your computer can be off, and it hands back a pull request or merge request. A session can start in the IDE, the Kiro CLI, or Kiro Web in a browser, and continue from any of them. An iOS app in early access through TestFlight follows and steers cloud sessions; it doesn't control a session running in the IDE on your computer. Kiro Web can also run scheduled automations, such as dependency updates, without anyone starting them.
Which models can Kiro use, and how are they billed?
Claude, OpenAI's GPT-5.6 family, and open-weight models from DeepSeek, Qwen, GLM, and MiniMax, all from Kiro's picker and all paid in Kiro credits rather than separate API keys. Each model has a multiplier against the Auto setting: Qwen3 Coder Next costs 0.05 times, Claude Sonnet 5.5 1.3 times, GPT-5.6 Sol 4.4 times. Credits are used fractionally per request, so a long spec task costs more than a quick edit, and they reset each month without rolling over. Some models are offered only in certain regions. The Pricing section has the plans.
How is Kiro different from Cursor?
Both are AI IDEs built on VS Code's open-source core, with a CLI and cloud agents that open pull requests. Kiro plans before it codes: a feature becomes requirements, a design, and tasks that you approve, which suits larger features and teams that review plans. Cursor leans on fast Tab completions and chat or agent edits, with its own Composer model. Kiro pays for every model in credits and includes open-weight models on its free plan, while Cursor's paid plan leads with Grok and frontier models.
Can I use Claude Code or Codex inside Kiro?
Yes. The official Claude Code and Codex extensions are published on Open VSX, the registry Kiro installs from, and any agent CLI runs in the integrated terminal. Each needs its own subscription, since Kiro's credits only cover Kiro's agent. They don't follow Kiro's spec workflow, but the requirements, design, and task files are plain Markdown in the repository, so another agent can read them as context and work through a task from the same plan.
Should every change go through a spec?
No. A spec costs time and credits to write and approve, which pays off for a feature that touches several files, needs agreement on behaviour, or will be reviewed by someone else. A one-line fix or a small refactor is faster in chat, and Quick Spec sits in between, writing all three documents without stopping for approval. The spec files stay in the repository, so they're worth keeping current: a design that no longer matches the code misleads the next person, or agent, that reads it.
Stacks Related to Kiro Dev Stack
Paseo Dev Stack
Developer
Run Claude Code, Codex, and other coding agents on your own machine with Paseo and steer them from any device.
Claude Code Dev Stack
DeveloperClaude Code in the terminal, desktop app, or web, paired with Claude and the add-ons to run it anywhere.
T3 Code Dev Stack
Developer
One open-source app for Claude Code, Codex, OpenCode, and other coding agents, on desktop and mobile.
Conductor Dev Stack
Developer
Run Claude Code, Codex, and Cursor's agent in parallel on your Mac with Conductor, each in its own workspace.
Scores
Popularity3/5
AWS backing and a broad set of surfaces make Kiro a well-known name among AI IDEs, though its user base trails Cursor and VS Code.
Learning Curve2/5
Anyone who has used VS Code can work in chat on day one; writing and approving specs, steering files, and hooks is a workflow that takes a few projects to get right.
Flexibility4/5
Six model families in one picker, Open VSX extensions, MCP, hooks, and powers, plus optional third-party agents, within a closed-source editor whose models are reached only through Kiro's credits.
Performance4/5
Parallel task waves and cloud sessions move long work off the local machine, though spec generation adds steps before any code appears and results track the chosen model.
Portability3/5
Projects stay plain files and git history, and the specs are Markdown any agent can read, but models, credits, and cloud sessions are tied to a Kiro account.
Tools in the Kiro Dev Stack Stack
Kiro Dev Stack Pricing
Kiro's subscription is the main cost, and its credits pay for every model in the picker. The free plan has 50 credits a month with open-weight models and Claude Sonnet 4.5; Pro is $20/month for 1,000 credits, up to Power at $200 for 10,000, with extra credits at $0.04. Cloud sessions need a paid plan. A separate agent like Claude Code or Codex needs its own plan. GitHub and GitLab are free for individuals.
Free: 50 credits; Pro $20 (1,000), Pro+ $40 (2,000), Pro Max $100 (5,000), Power $200 (10,000) per user; extra credits $0.04; Enterprise on request.
Free tier covers individual devs and small teams; GitHub Team is $4/mo, GitLab Premium is $29/mo for more seats.
Claude Code starts at $20/mo (no free tier); Codex has a free tier, with Plus at $20/mo; both go to $200/mo. OpenCode is free and bills per token.
Docker Desktop is free for personal use; Docker Pro adds team features from $5/mo.
GitHub Actions and GitLab CI/CD both include free monthly minutes before paid tiers kick in.