Kiro turns prompts into requirements, designs and task lists, implements them with parallel agents and checks code with property-based tests. It runs as a desktop IDE, CLI, web app and iOS app, with a 50-credit free tier and paid plans from $20 per user per month.
Used this tool? Rate it
Used this tool? Rate it
Kiro is an AI coding agent that ships as a desktop IDE, a command-line interface, a web app, an iOS app and Kiro Crew, all built around turning prompts into executable specs and working code. Kiro turns a prompt into requirements, an architectural design and a sequenced task list, then implements those tasks with parallel agents.
Kiro is built and operated by a small team within AWS, and the Kiro IDE and Kiro CLI are each licensed as AWS Content under the AWS Customer Agreement. You do not need an AWS account to use it: sign-in works with GitHub, Google, AWS Builder ID or AWS IAM Identity Center. Kiro was unveiled as a preview in July 2025 and is now sold as credit-based plans next to a perpetual free tier.
The IDE is based on Code OSS, so VS Code settings, themes and Open VSX compatible plugins can be imported during onboarding. Other, unrelated businesses also use the Kiro name on different domains; this page covers the AI development tool at kiro.dev.
Specs capture what should be built, steering files and powers give the agent context, hooks automate follow-up work, and property-based tests check the result. The IDE, CLI, web app and iOS app are front ends to one shared agent harness, so a steering rule configured once behaves the same on every surface where the feature is available.
Before code is written, Kiro checks requirements for contradictions and gaps using automated reasoning techniques. Kiro then extracts properties from EARS-formatted requirements and generates hundreds or thousands of random test cases when you choose to run them. Property-based tests are optional by default, and the correctness feature is available in the IDE only, while Analyze Requirements also works in the CLI.
By default, prompts go to Auto, a router that mixes frontier models with specialized models for intent detection and caching. You can instead choose OpenAI's GPT-5.6 tiers (Sol, Terra and Luna), Anthropic's Claude models (Opus 5, Opus 4.8, Sonnet 5, Sonnet 4.6, Sonnet 4.5 and Haiku 4.5), or open weight models such as DeepSeek 3.2, MiniMax M2.5, GLM-5 and Qwen3 Coder Next. The model choice changes the cost of the same work: Sonnet models use a 1.3x credit multiplier versus 2.2x for Opus, and Qwen3 Coder Next has a 0.05x multiplier, the lowest in the list.
kiro-cli chat --no-interactive with your instruction, and grant tool permissions up front with --trust-all-tools or a narrower --trust-tools list, because nobody is present to approve tool calls.Kiro is built for professional developers on software teams, businesses and enterprises, with plans sold per user for teams up to 500 people. Choosing it makes most sense in these situations:
Kiro fits less well for developers who want browser or phone access without paying, since the web and iOS apps require a paid plan, or for anyone who needs the smallest possible monthly bill under heavy agent use.
Enterprise profiles are hosted in US East (N. Virginia), Europe (Frankfurt) or AWS GovCloud (US), and SSO works through Okta, Entra ID and AWS IAM Identity Center.
Kiro pricing is credit-based; prices are in USD and exclude taxes, and paid plans are priced per user.
| Plan | Price | Credits per month | Models | Add-on credits |
|---|---|---|---|---|
| Kiro Free | $0 per month | 50 | Open weight models and Claude Sonnet 4.5 | Not available |
| Kiro Pro | $20 per user per month | 1,000 | Premium models | $0.04 per credit |
| Kiro Pro+ | $40 per user per month | 2,000 | Premium models | $0.04 per credit |
| Kiro Pro Max | $100 per user per month | 5,000 | Premium models | $0.04 per credit |
| Kiro Power | $200 per user per month | 10,000 | Premium models | $0.04 per credit |
A credit is a unit of work, not a request: simple prompts can use less than 1 credit, a spec task typically costs more, and usage is metered down to 0.01 credits. Spec refinement, task execution and agent hook runs all consume credits. The homepage describes the credit-based model as having "no daily or weekly rate limits" without naming a plan, while the pricing page footnote says Free-tier model access "is subject to rate limits"; the two pages are inconsistent on this point.
Plan credits reset at the start of each billing month and do not roll over. Add-on credits are the exception: packs range from $5 (125 credits) to $100, up to five packs can be held at a time, and unused add-on credits expire 12 months after purchase. When plan and add-on credits both run out, requests pause until the next month or a new purchase.
The first upgrade from Free comes with a $20 credit toward the subscription when you sign in with a social login or AWS Builder ID. Billing runs on calendar months: a first upgrade mid-month is prorated, a mid-month downgrade still costs the full fee for the current month, and cancelling keeps the paid plan until the month ends before switching to Kiro Free. The billing pages reviewed list refund requests as a billing-support topic but state no refund window. Pricing in AWS GovCloud (US) is approximately 20% higher and has no free tier.
A product-launch listing for Kiro places Cursor, Claude Code and Kilo Code among its similar products. Two of them compare usefully:
Property-based testing provides evidence of correctness, not a proof, and it is not formal verification. A property that is too weak or states the wrong invariant will pass while the real behavior is still wrong, and requirements that depend on external services may need mocking or example-based tests instead.
On the Free tier and on individual paid plans (social login or AWS Builder ID), Kiro may use your questions, other inputs, and the responses and code it generates for service improvement, including model training. This collection is on by default; in the IDE you can opt out by unchecking Content Collection for Service Improvement under Settings, Application, Telemetry and Content. Content from enterprise users is not used for service improvement, and they are automatically opted out of telemetry and content collection. The enterprise page states more broadly that content is not used for training foundation models on Pro, Pro+, Pro Max or Power subscriptions, which does not match the narrower data-protection page for individual subscribers; the two pages are inconsistent.
Retention also differs by plan. Free-tier inputs may be stored for up to 60 days for abuse detection even after opting out, and classifier-flagged traffic to OpenAI GPT models is retained for up to 30 days. Content from Free and individual users is stored in the US East (N. Virginia) Region. Paid subscribers receive an output indemnity under the AWS Service Terms.
Enabling network access for the Kiro Web sandbox exposes the environment to prompt injection, extraction of code and secrets, and malware; Kiro describes connections-only access as the most secure level and recommends it for tasks without external dependencies.
The perpetual Kiro Free tier includes 50 credits with access to open weight models and Claude Sonnet 4.5, subject to rate limits. Kiro Web and the iOS app require a paid plan.
Free-tier users and individual subscribers who use a social login or AWS Builder ID can have prompts and generated code used for service improvement, including model training, unless they opt out. Enterprise users signing in through IAM Identity Center or an external identity provider are excluded.
No. You can sign in with GitHub, Google or AWS Builder ID.
On Free, requests pause until the monthly reset or an upgrade. On paid plans, you can buy add-on credits at $0.04 each or move to a higher tier.
Auto is the default. Paid plans can also pick Claude models such as Opus 5, OpenAI's GPT-5.6 tiers or open weight models; the Free tier is limited to Claude Sonnet 4.5 and open weight models.