Kiro Crew vs Kiro CLI: workspace and remote workflows

Johannes interviews Darko about Kiro Crew, an AI agent workspace built around Kiro CLI. Darko explains how it uses ACP to communicate with the CLI and adds session management, scheduled tasks and agent orchestration. The discussion compares its workflow with the CLI and IDE, then considers how much control it offers alongside Kiro Web.

Darko demonstrates instances on his laptop and a Linux ARM workstation. He says the workspace can also run on a Raspberry Pi or remote server. For self-hosted access, he describes a web interface restricted to localhost by default and an SSH tunnel to port 5476. The workspace still connects to Kiro’s service; the interview does not demonstrate local model inference. Crew is described as open source under Apache 2.0, while the current CLI is not.

The coding assistant demo asks Crew to create an issue and pull request for light mode on pingdarko.com. Alongside it, Darko explains a vector knowledge base and automatic lessons generated from past sessions during idle time. Apps extend the workspace with repository tools and file browsing.

The interview also covers command approvals and security policies. At recording time, Crew supported only Kiro CLI, and agent plugin support was not yet available according to Darko. He warns that autonomous loops may consume more credits, but says he has not measured usage against the CLI.