KiroCrew - tame your agents and manage them

Taming your local and cloud agents with KiroCrew

While I was in Seattle, attending the AWS Heroes Summit 2026, a team at AWS has announced KiroCrew. KiroCrew is a new OpenSource project that has been built inside AWS.

The readme says:

Open-source personal AI agent that runs on your own machine. Chat from Slack, a web dashboard, or the CLI; run multi-step tasks unattended; schedule cron jobs; persist memory across sessions.

KiroCrew orchestrates the kiro-cli agent over the Agent Client Protocol (ACP), adding multi-session management, persistent memory, scheduling, and a web UI on top of it.

But what does that mean?

Orchestrating agents is the new problem

In a world of AI agents running “everywhere” - locally and in the cloud - the next challenge for builders is to “orchestrate agents” and optimize your harness.

The team at AWS seems to believe that orchestrating the agents, connecting them with each other and connecting tools and MCP servers is the next problem to solve.

You need to be able to efficiently switch between different sessions and contextes, with different connections to (remote) tools, knowledge and agents.

KiroCrew overview

This diagram is straight from the Open Source repository and shows the architecture of KiroCrew:

graph TB
    subgraph "User Surfaces"
        CLI[CLI<br/><code>kirocrew chat</code>]
        Slack[Slack DM<br/>Socket Mode]
        Dashboard[Web Dashboard<br/>React SPA]
        Desktop[Desktop App<br/>Electron]
    end

    subgraph "KiroCrew Gateway"
        GW[Gateway Process<br/><i>Python / aiohttp</i>]
    end

    subgraph "Agent Backend"
        KC[kiro-cli<br/>ACP over stdio]
        LLM[LLM Provider<br/><i>Claude via Amazon</i>]
        MCP[MCP Servers<br/><i>tools</i>]
    end

    CLI --> GW
    Slack --> GW
    Dashboard --> GW
    Desktop --> Dashboard

    GW --> KC
    KC --> LLM
    KC --> MCP

In short you can say that KiroCrew manages sessions of the kiro-cli which are connected over ACP (Agent Contenxt Protocol). This allows the UI to manage connections to Agents both locally and remotely. And this allows you to manage agents that run 24/7 if you put the Agent Backend on a remote backend (e.g. an EC2 instance…).

This page clearly shows the value add that KiroCrew gives over the plain kiro-cli. It also gives a lot of background on why KiroCrew was built and created inside AWS and why it seems to be getting traction right now, especially insight AWS.

It allows you to also connect and talk to your agents through Slack.

Trying it out

It seems to be simple to try KiroCrew but I need to admit that I wasn’t able to try it out myself yet. This has mainly the reason that I didn’t have enough time yet :-)

If you’ve tried it, please write a blog post about it and share it with me so I can link it here!

Interesting findings in the repository

AGENTS.md gives a tiny bit of insights into whow this project is built (using Kiro - and Claude Code) inside AWS and how the team seems to struggle to tame the agents from introducing AWS specific dependencies.

It’s the context that matters - and managing a non-AWS specific context on an AWS configured laptop seems to be challenging ;)

Kiro Crew at the Heroes summit

At the AWS Heroes Summit in Seattle there was a lot of discussion and hype around Kiro Crew and a lot of people where really interested to understand and see the product in action.

Have you tried KiroCrew yet? I’d love to hear — reach out to me on LinkedIn or through my contact form.