K

Kiro Crew

A persistent AI development workspace that runs on your hardware

Productivity Open Source The software itself is open source under Apache-2.0 and free to self-host; running it requires signing in to a Kiro account (device-code sign-in on first launch), and model requests go through kiro-cli, counting against your Kiro plan's usage. ★ 4.3 Visit Website →

About

Kiro Crew is an open-source (Apache-2.0) persistent development workspace from the Amazon Kiro team that runs on hardware you control—your Mac, a local container, or a remote Linux host. Unlike agents that stop when the chat closes, Kiro Crew is built around a long-running Gateway, so sessions, memory, schedules, and task checkpoints survive restarts. It is self-learning and self-evolving: corrections and task failures become durable lessons, and repeated patterns are synthesized into reusable skills. Work with it from the desktop app, the web dashboard (localhost:5476 by default), the CLI, or messaging surfaces like Slack, Discord, Telegram, Teams, Webex, WeCom, and WeChat—all against the same runtime and state. Multi-step tasks run unattended: it plans steps, executes, validates results, retries failures, and resumes from checkpoints. Recurring jobs run on your schedule and heartbeats monitor systems until something needs attention. Kiro Crew Apps package a purpose-built interface together with agents, skills, schedules, integrations, and backend services. Underneath it drives kiro-cli over ACP, installing it on first launch if needed and guiding Kiro device-code sign-in.

Key Features

  • Long-running Gateway: sessions, memory, schedules, and task checkpoints survive restarts
  • Self-learning: corrections and task failures become durable lessons
  • Self-evolving: synthesizes reusable skills from repeated patterns, inspectable and editable
  • Unattended long-running tasks: plan, execute, validate, retry, resume from checkpoints
  • Scheduled, proactive, and reactive work—cron jobs, AutoNudge goal loops, authenticated webhooks and messaging events
  • Subagent delegation: fan out isolated parallel sessions and merge results back
  • Same state from every surface—desktop app, web dashboard, CLI, and messaging
  • Messaging channels—Slack, Discord, Telegram, Teams, Webex, WeCom, WeChat
  • Runs where you choose—local Mac, Docker container, or a remote host (desktop app can tunnel to a remote Gateway over SSH)
  • Kiro Crew Apps: ship an interface plus agents, skills, schedules, integrations, and backend services
  • Extensible tooling—MCP servers, markdown skills, and hooks, with built-in kirocrew-core / kirocrew-cron
  • Visible execution: tool calls, subagent progress, context usage, approvals, memory, and logs in the dashboard
  • Defense in depth: tool approvals, Linux/macOS OS sandboxing, sensitive-path checks, credential redaction, deny rules, audit events, and governance profiles
  • Local by default: the dashboard binds to loopback, and remote access requires token authentication
  • Multiple install paths and channels—desktop app (macOS DMG / Linux AppImage), one-line CLI install, multi-arch Docker image, each on Stable / Insider / Nightly
Compare DeepSeek Harness vs Kiro Crew: two open-source agent routes compared

Alternatives to Kiro Crew