Kiro Crew
by Amazon Web Services (AWS)
Open-source workspace for coding agents that keep working after you close the laptop
Kiro Crew is AWS's open-source orchestration workspace for asynchronous coding agents, released under Apache 2.0 in August 2026. It lets engineering teams hand long-running work — migrations, dependency upgrades, ticket triage, incident investigation, PR monitoring — to multiple agents that share persistent memory, run on a schedule and check in for approval, all on infrastructure the team controls.
Kiro Crew is AWS's open-source orchestration workspace for asynchronous coding agents, released under the Apache 2.0 licence on 4 August 2026. It was developed inside Amazon under the name MeshClaw, where AWS says more than 39,000 developers adopted it before the public release. Where most AI coding tools are synchronous — a developer prompts, waits and reviews — Crew is built for work that outlives a single session: incident investigation, ticket triage, dependency upgrades, large migrations and pull-request monitoring, run unattended with checkpoints, validation and automatic retries. It is structured around a Gateway that separates the agent runtime from the interfaces, so the same crew is reachable from a desktop app on macOS, Windows and Linux, a web dashboard, a terminal UI, a CLI and messaging surfaces including Slack, Discord, Telegram and Microsoft Teams. Agents share persistent memory that survives Gateway restarts, synthesise repeated patterns into reusable and inspectable Skills, run scheduled jobs and webhook triggers, and delegate to subagents that work in parallel and report back. Purpose-built Apps such as Issue Radar, DevFleets and Task Runner package common workflows so they do not have to be re-prompted each time. Tooling is wired through the Agent Client Protocol and the Model Context Protocol. Security is deliberately conservative: OS-level sandboxing, deny-by-default commands with 137 bundled deny patterns, path restrictions, credential redaction, interactive approvals and audit logging, with the dashboard bound to localhost by default. It runs entirely on infrastructure the customer controls — laptop, container or VM — with no AWS account or AWS-managed control plane required.
A platform engineering or DevOps lead who already runs Kiro and wants repetitive, long-running engineering work — upgrades, triage, migrations — done unattended under approval gates and audit logging rather than by individual developers babysitting prompts.
Multi-session engineering work runs to completion without a human holding the context, inside a sandbox and audit trail the organisation controls.
At a Glance
- Category
- Agent Development
- Pricing
- Free, Subscription, Usage-based
- Target Market
- CTOs, VPs of Engineering, Platform Engineering Teams, DevOps and SRE Leads, Enterprise Developers
- Deployment
- Open-source, Self-hosted, Hybrid
- Founded
- 2006
- Headquarters
- Seattle, United States
- Team Size
- 500+
- Customers
- 39,000+ internal Amazon developers reported pre-release; ~3.5k GitHub stars since open-sourcing
Key Features
- ✓Asynchronous multi-agent orchestration
Assign work to several agents that run concurrently and unattended, returning results rather than requiring live supervision.
- ✓Persistent memory across sessions
Conversations and project context survive Gateway restarts, with searchable history, so recurring work does not cold-start.
- ✓Self-evolving Skills
Repeated patterns and past corrections are synthesised into reusable, inspectable skills that shape future agent behaviour.
- ✓Apps for packaged workflows
Purpose-built interfaces such as Issue Radar, DevFleets and Task Runner run workflows without chat instructions each time.
- ✓Scheduling, heartbeats and webhooks
Recurring jobs and event triggers let agents act outside active sessions, which is what makes unattended operation possible.
- ✓Defence-in-depth sandboxing
OS-level sandboxing, 137 bundled deny patterns, path restrictions, credential redaction and audit logging are on by default.
- ✓Open protocols and open licence
Apache 2.0, built on Agent Client Protocol and Model Context Protocol, deployable via Docker or source on Python 3.10+ and Node 22+.
Capabilities
Use Cases
- •Unattended dependency upgrades
A scheduled crew upgrades libraries across repositories overnight, validates builds and opens pull requests for human review.
- •Incident investigation and triage
On a webhook from the alerting stack, agents gather logs and context so the on-call engineer starts from a summary.
- •Large codebase migrations
Long-running migration work is checkpointed and retried automatically, surviving restarts instead of dying with the session.
- •Pull-request monitoring at scale
Agents watch open PRs across many repositories, flag stale or failing ones and report into Slack or Teams.
- •Governed replacement for shadow agent use
Individually-run developer agents move into one shared workspace with approval gates, logging and visible tool calls.
Ideal For
Best For
- ✓Running long migrations and dependency upgrades unattended with checkpoints and automatic retries
- ✓Automating ticket triage and incident investigation for DevOps and SRE teams
- ✓Continuous pull-request monitoring and review across many repositories
- ✓Consolidating individually-run coding agents into one governed workspace with approval gates and audit logs
- ✓Self-hosting an agent orchestration layer with no vendor control plane and no AWS account requirement
Not Ideal For
- ✗Teams standardised on Claude Code, Codex, Cursor or Devin — Crew runs on the proprietary Kiro CLI at launch, so those agents need custom connectors built and validated first
- ✗Organisations with no cost-measurement framework for AI agents; practitioners report Crew burns tokens noticeably faster than Kiro CLI alone, and parallel subagents multiply that
- ✗Solo developers doing short synchronous edits, for whom a plain coding assistant is simpler and cheaper than a Gateway, dashboard and scheduler
- ✗Enterprises that have not yet set policy on least-privilege access, memory retention and auditability for persistent agents — analysts flag that as a prerequisite, not a follow-up
Integrations
Deployment
Market & Ratings
39,000+ internal Amazon developers reported pre-release; ~3.5k GitHub stars since open-sourcing
Market Analysis
Pros
- ✓Genuinely open source under Apache 2.0 and deployable entirely inside customer infrastructure, so there is no control-plane lock-in or data-egress question
- ✓Real internal production history at Amazon scale before release, which is rare for a newly open-sourced agent framework
- ✓Governance features enterprises actually ask for — approval gates, audit logging, visible plans and tool calls — instead of leaving agents as per-developer shadow IT
- ✓Broad surface coverage: desktop, web dashboard, CLI, TUI and chat, so the same crew can be driven from wherever the team already works
- ✓Active project with over 4,700 commits and a documented governance and contribution model
Cons
- ✗Runs on the proprietary Kiro CLI at launch, so teams on Claude Code, Codex or Devin must build and validate their own connectors — analysts call this the biggest adoption blocker
- ✗Practitioners report noticeably higher token consumption than Kiro CLI alone, and parallel subagents make cost hard to predict
- ✗686 open issues against 3.5k stars is a heavy backlog for a month-old public project, so expect rough edges
- ✗Human approval workflows, which are what make it safe, also throttle the very automation benefit being bought on long workflows
- ✗Analysts note most enterprises still lack a framework to measure per-agent cost, making the business case for multi-agent swarms hard to defend internally
- ✗Persistent agent memory demands least-privilege and retention policy work before deployment, which is organisational effort the tool cannot supply
Pricing
Kiro Crew (open source)
$0
- ✓Apache 2.0 licence
- ✓Self-hosted on laptop, container or VM
- ✓No AWS account required
- ✓Desktop, web, CLI and TUI interfaces
Kiro Free
$0
- ✓50 credits per month
- ✓Open-weight models and Claude Sonnet 4.5
- ✓Not available in AWS GovCloud
Kiro Pro
From $20/mo
- ✓1,000 credits per user
- ✓Premium models
- ✓Add-on credits at $0.04 each
Kiro Pro Max
From $100/mo
- ✓5,000 credits per user
- ✓Premium models
- ✓Add-on credits at $0.04 each
Kiro Power
From $200/mo
- ✓10,000 credits per user
- ✓Premium models
- ✓Add-on credits at $0.04 each
The Crew software itself is free and Apache 2.0, but at launch it drives the proprietary Kiro CLI, so the real cost is a Kiro seat metered in credits: Free at 50 credits, Pro $20/user/month for 1,000, Pro+ $40 for 2,000, Pro Max $100 for 5,000 and Power $200 for 10,000, with overage add-on credits at $0.04 each. Credits are metered to two decimal places and do not roll over month to month, and model choice changes the burn rate — the same task costs about 1.3x more on Sonnet 4.6 than on Auto. Budget carefully for Crew specifically: practitioners report it consumes tokens faster than Kiro CLI alone because subagents run in parallel. AWS GovCloud pricing runs roughly 20 percent higher and has no free tier, and centralised billing, SSO, usage analytics and custom security controls are enterprise items handled through sales.
Security & Compliance
Connect
Sources
This page was written from 6 sources, 4 on domains other than kiro.dev.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Naïve
Autonomous company infrastructure — compute, payments, identity and incorporation for AI agents behind one API
Microsoft Agent Lightning
MIT-licensed framework that trains any AI agent with reinforcement learning without changing its code
Keenable
Independent web search API for AI agents — 100B+ document index, sub-250ms p95
LangSmith
Framework-agnostic platform to observe, evaluate, deploy and continuously improve production AI agents