Kiro Crew Review (2026): Pricing, Features & Honest Verdict

Kiro Crew is an open source agentic development workspace that keeps context, lessons and skills alive between sessions instead of starting every task cold. Our verdict: the persistent memory is the actual product, and it earns its keep. Best for: engineering teams already running AI coding agents daily. Price: free under Apache 2.0, but it runs on a Kiro plan (free tier available, Pro $20/mo). Rating: 8/10.

What is Kiro Crew?

Kiro Crew is a workspace that sits between you and a fleet of AI agents. You give it work, it keeps a memory of how your tools, repos and preferences fit together, and the next time you come back it picks up where the last session ended. It ships as a desktop app, a web dashboard, a CLI, and a bot you can talk to from Slack, Discord or Telegram. It launched on Product Hunt on 5 August 2026 and landed in the top ten on its first day.

The origin story is more interesting than most. Crew started life inside Amazon as an internal tool called MeshClaw, built by three engineers: Bolin Chen, Zejiang Guo and Zezhen Xu. It spread through Amazon’s internal developer population on its own merits, reaching over 39,000 users and close to 500 contributors in roughly six months, and then got open sourced under Apache 2.0 as a second product line alongside Kiro’s spec-driven IDE. That matters, because a tool with six months of hard internal use behind it is a very different proposition to a launch-day repo.

What are Kiro Crew’s key features?

Persistent memory with inspectable lessons and skills

This is the headline and it is not marketing. Crew builds a knowledge graph with vector embeddings over your context, then extracts repeated patterns into what it calls skills. A skill is a reusable workflow the agent learned from watching you do the same thing three times. The important design choice is that both lessons and skills are inspectable: you can read them, edit them, and delete the ones that are wrong. Most agent memory is a black box that quietly poisons future runs. This one is not.

Apps instead of chat prompts

Crew’s Apps are purpose-built interfaces for jobs you repeat: issue triage, feature flag management, task execution. You are not typing a paragraph of instructions every morning. You open the App, it already knows the shape of the job, and it runs. You can build your own in TypeScript or Python. This is the single biggest usability difference between Crew and a bare terminal agent, and it is the part that will decide whether a team keeps using it after week two.

Autonomous triggers and parallel agents

Crew runs on cron schedules, webhooks and heartbeat monitors, so work happens while you are asleep. Migrations, incident investigations and recurring reviews are the use cases the team leads with. It also runs multiple specialised agents in parallel, each with an isolated context, which is the correct architecture: one agent with everything crammed into its window degrades fast, and separate agents with separate contexts do not.

Security that was clearly designed for a large company

OS-level sandboxing, command filtering, credential redaction and signed audit logs. Every action streams in real time through the Agent Client Protocol, so you can watch what an agent is doing rather than reading the wreckage afterwards. You can tell this came out of Amazon rather than a weekend hackathon. If you have ever had to explain to a security team why an AI agent has a shell on a production box, this is the feature list you want to hand them.

Four interfaces, one workspace

Desktop app (Electron, no extra dependencies), web dashboard, TUI, and chat integrations. Downloads come from GitHub releases or the web dashboard. The chat route is underrated: being able to fire off a migration from a Slack thread and have the audit log land back in the same thread removes a lot of context switching.

How does Kiro Crew compare to running agents yourself?

Most teams doing serious agent work have already assembled something like this out of parts: a repo of prompts, a few shell scripts, a cron job, and a Slack webhook someone wrote in a hurry. Crew’s pitch is that the assembly is the hard part and you are not being paid to do it.

Where it genuinely wins is state. A home-grown setup restarts from zero every session because nobody builds a knowledge graph for an internal tool. Where it loses is control. Your scripts do exactly what you wrote, and Crew does what it learned, which is better on average and worse on the occasions it is wrong. The inspectable skills list is the mitigation, and we would use it aggressively in the first month rather than trusting the defaults.

One more honest caveat: this is an Electron desktop app coordinating LLM calls. It is not lightweight, and the hardware bill is yours whether you run it on a laptop or a cloud instance.

How much does Kiro Crew cost?

Crew itself is free. Apache 2.0, no proprietary components, no paid edition of the workspace. What it needs is a Kiro account, because model usage counts against your Kiro plan rather than a separate Crew subscription.

Kiro’s ladder starts at a free tier with 50 credits a month and open weight models plus Claude Sonnet 4.5. Pro is $20 a month for 1,000 credits, Pro+ is $40 for 2,000, Pro Max is $100 for 5,000, and Power is $200 for 10,000. Every paid tier adds premium models and add-on credits at $0.04 each, which is also the overage rate. Team plans cost the same per user and add consolidated billing, usage analytics, SSO through AWS IAM Identity Center and an organisational dashboard. New users upgrading through social login or an AWS Builder ID get $20 credited toward a first paid plan.

Against the market that is fair rather than cheap. Devin runs a free tier, $20 Pro and $200 Max, with Teams at $80 a month plus $40 per developer seat. Cursor’s Pro plan is $20 a month. Kiro sits exactly on that line, and the difference is that Crew is the only one of the three where the orchestration layer itself is open source and self-hostable in the sense that you own the code and the hardware.

The honest framing: “open source” here means you can read and fork the workspace, not that you can run it for nothing. The 50 free credits will show you whether it fits. They will not run a migration.

Who is Kiro Crew best for?

Use Kiro Crew if your team already runs AI agents most days and the cold start is the thing that annoys you. It is aimed squarely at people who have felt the specific pain of re-explaining their repo layout to an agent for the fortieth time. It also suits teams with a security function that asks hard questions, because the sandboxing and signed audit logs give you real answers.

Use it if you want autonomous work: overnight migrations, scheduled dependency reviews, incident triage that starts before you wake up. That is where scheduled triggers plus persistent memory compound into something a chat window cannot match.

Skip Kiro Crew if you are a solo developer who opens an agent twice a week. The setup cost will not pay back. Skip it if you need everything on hardware you control with no external model calls, because the credit system routes through Kiro. And skip it for now if you need a mature, stable tool with three years of production hardening: this is six months of internal Amazon use plus a public launch, which is a good start and not the same thing.

Best Kiro Crew alternatives

OpenHands is the closest philosophical match. MIT licensed, free to run locally, and the individual cloud tier is free too with bring-your-own-key or at-cost model access. Enterprise is custom priced with self-hosting in your own VPC. Pick OpenHands if licence purity matters more to you than persistent memory, because that is the trade: OpenHands is more permissively licensed, Crew remembers more.

Devin is the managed opposite. Free tier, $20 Pro, $200 Max, and Teams at $80 a month plus $40 per full developer seat. You get a hosted agent that needs no local infrastructure at all, and you give up the ability to read the source or run it yourself. Pick Devin if you want zero operational surface.

Cursor at $20 a month for Pro solves a smaller problem well. It is an editor with excellent agent integration rather than an autonomous workspace, so it will not run a cron job at 3am. Pick Cursor if the work you want automated still happens inside a file you have open.

Final verdict: is Kiro Crew worth it?

Yes, if you are in the target group. The persistent memory is the feature, and unlike most memory claims in this category it is built on a knowledge graph with skills you can actually open and correct. The Apps concept solves the second real problem, which is that nobody wants to write a paragraph of instructions to do a task they do weekly. Six months of internal Amazon use and 500 contributors is meaningful validation that this survived contact with real engineers before it reached you.

The caution is the pricing frame. Crew is genuinely open source, but the operating cost is a Kiro plan, and if your team lands on Pro that is $20 per person per month before you count overage credits at four cents each. Budget for the credits, not the licence. If your team runs agents daily and keeps losing context between sessions, this is the strongest answer we have seen to that specific problem, and it is worth the trial. If agents are still an occasional experiment for you, wait a quarter and let the project mature.