Kiro Crew

Product Introduction

1. Definition: Kiro Crew is an open-source, persistent AI agent orchestration platform and development workspace. Technically, it is a self-hosted application that runs on the Kiro CLI and uses the open Agent Client Protocol (ACP) to manage multiple AI agents that operate across sessions, tools, and workflows.
2. Core Value Proposition: It exists to eliminate the "cold start" problem in AI-assisted development by providing a persistent, context-aware workspace. Its primary value is enabling developers to build a crew of autonomous agents that learn from interactions, remember project context, and execute scheduled, triggered, and long-running tasks unattended, thereby increasing engineering productivity and automating repetitive workflows.

Main Features

1. Persistent Memory & Self-Improvement: Kiro Crew maintains a durable memory system that survives application restarts and sessions. It synthesizes user corrections into Markdown-based "lessons" and repeated workflows into reusable "skills." Context is stored in a knowledge graph backed by vector embeddings and full-text search (likely using technologies like pgvector or similar), enabling precise, semantic retrieval instead of full-context re-reading.
2. Unattended Agent Orchestration: The platform enables background and scheduled work through multiple triggers. It supports timezone-aware cron jobs, authenticated webhook endpoints, and heartbeat monitors that watch state changes in PRs, deployments, and CI/CD pipelines. This allows agents to autonomously scan for issues, monitor codebase drift, and react to CI/CD events without user initiation.
3. Defense-in-Depth Security Model: Kiro Crew implements a 7-layer security architecture for safe, autonomous code execution. Key layers include OS-level process sandboxing, a deny-by-default command policy, bash pattern blocking, input/output validation for MCP (Model Context Protocol) servers, sensitive filesystem path protection, automatic credential redaction in logs, and cryptographically signed audit logs (SEL). Being open-source, every layer is auditable.
4. Extensible App Framework: Developers can build purpose-built interfaces, called "Apps," using a TypeScript/React or Python SDK. These Apps can wrap agents, skills, and schedules into custom UIs, register MCP tools, subscribe to agent events, and persist state. Example built-in Apps include a "Research Lab" for multi-agent research and a "Code Review Sage" that conducts per-file reviews.
5. Multi-Platform Interface: The system is accessible through a responsive web dashboard, a full-featured CLI (kirocrew chat, run, cron), a standalone Electron desktop app, and chat integrations (Slack, Discord, Telegram). It can also manage remote instances over SSH, offering deployment flexibility.

Problems Solved

1. Pain Point: Context loss and repetitive instruction between AI coding sessions. Developers often re-explain project architecture, coding patterns, and past decisions to AI assistants, wasting time and breaking flow.
2. Target Audience: Software engineers, DevOps professionals, and engineering managers who use AI coding tools extensively. Specifically, developers working on complex, long-lived projects who need to automate code review, dependency updates, backlog triage, and CI/CD monitoring.
3. Use Cases: Automatically reviewing pull requests by spawning sub-agents per file; monitoring production deployments and triggering rollback procedures; running nightly dependency audits and security scans; triaging a GitHub issue backlog for stale or actionable items; maintaining internal documentation by syncing code changes with a knowledge base.

Unique Advantages

1. Differentiation: Unlike single-session AI coding tools (e.g., Cursor, GitHub Copilot) or general-purpose agent platforms, Kiro Crew is specifically engineered for persistent, multi-agent development work with deep CI/CD and codebase integration. It prioritizes inspectability (open-source, live activity views, audit logs) and security (sandboxing, redaction) over being a black-box automation tool.
2. Key Innovation: Its additive, interoperable design built on open standards. It uses the Kiro CLI and ACP, is compatible with skills and MCP servers from other platforms (like OpenClaw/Hermes), and reads existing .kiro configurations. This allows teams to adopt its persistent orchestration capabilities without replacing their existing AI toolchain, reducing lock-in and migration cost.

Frequently Asked Questions (FAQ)

1. How does Kiro Crew handle security and safety for autonomous code execution? Kiro Crew employs a 7-layer defense-in-depth strategy including OS sandboxing, denied command lists, bash pattern blocking, and credential redaction. Every action is logged in a signed audit log, and because it's open-source, users can verify and audit every security control themselves.
2. Can Kiro Crew run completely offline or with local models? Kiro Crew runs on your own infrastructure (laptop, server, cloud VM). While the orchestration platform is local, the agents it coordinates typically require inference from cloud-based LLMs via your Kiro account. Its use of the open ACP means it could, in principle, orchestrate agents connected to local models if an ACP-compatible runtime supports them.
3. What is the cost structure for using Kiro Crew? The Kiro Crew software itself is free and open-source (Apache 2.0). However, it requires a Kiro account and plan to provide the AI model inference that powers the agents. Usage from Kiro Crew counts against your standard Kiro plan credits, the same as using the Kiro IDE or CLI. You also provide the compute hardware it runs on.
4. How does the persistent memory and knowledge graph actually work? Corrections and workflows are saved as editable Markdown files (lessons & skills). Broader project context and decisions are stored in a knowledge graph that uses vector embeddings for semantic search and traditional full-text search. This allows agents to query for relevant past information efficiently without costly re-processing of entire conversation histories.
5. Is Kiro Crew a replacement for CI/CD tools like Jenkins or GitHub Actions? No, it is a complementary orchestration layer. It reacts to events from your CI/CD pipelines (e.g., a failed build webhook) and can execute complex, context-aware follow-up actions using AI agents. It handles the "what to do next" reasoning, while traditional CI/CD handles the deterministic "how to run" steps.