Kiro Crew Documentation

September 6, 2026 · View on GitHub

Kiro Crew is a personal, autonomous AI agent that runs locally on your own machine. It is powered by kiro-cli (KiroACP) and reaches tools over the Model Context Protocol (MCP). Everything below is the reference for the features you can reach from the dashboard, the CLI, or a connected messaging channel.

Quick Start

Install the prebuilt, signed wheel:

curl -fsSL https://download.crew.kiro.dev/cli.sh | sh

Then start the gateway and open the dashboard:

kirocrew gateway     # → http://localhost:5476

kiro-cli must be installed, on your PATH, and logged in. See Getting Started for the source install, the pip channel index, first-time setup, and connecting messaging channels.

Core Capabilities

Capability	Description
Cron Jobs	Schedule recurring tasks, e.g. "every weekday at 9am give me a pipeline briefing"
Subagents	Spawn parallel background workers for fan-out research and multi-package work
Dynamic Sub-Agent Sizing	Auto-size the concurrent sub-agent cap from host memory/CPU and a learned per-agent cost
Memory	Persistent preferences, project context, and learned corrections across sessions, plus per-session persistent / incognito / temporary memory modes
Task Runner	Autonomous multi-step execution from spec files: hand it a task, walk away
Research Lab	Autonomous multi-cycle research campaigns with scoping, adaptive agent execution, and exportable reports
Dashboard	React web UI with multi-session chat, memory management, and live system metrics
Agent Questions	Let an agent pause mid-turn and ask you a clickable multiple-choice question
Chat Channels	DM-based chat with tool approval — Slack, Discord, Telegram, Teams, Webex, WeCom, WeChat, iMessage, WhatsApp, Feishu; per-channel capabilities in each guide
Agents	Switch between specialized agents per conversation, thread, or cron job
Skills	Drop-in markdown knowledge packs for domain-specific workflows
Dynamic Workflows	Multi-phase agent orchestration authored from a plain-language goal: watch a run, restart part of it from cache, save it for reuse
Artifacts	Save, version, and revert generated UI and documents so they outlive the chat scrollback
Monitor Loops	Keep one session checking something on an interval — a pull request, a CI run, a deployment — until an exit condition fires
Session Ledger	A durable per-session record of goal, phase, and next step that survives context compaction
Browser Control	Drive a real web page from the dashboard's Browser panel: navigate, snapshot, click, type, screenshot
Computer Use	Read and drive native desktop applications through the accessibility layer; opt-in and off by default

Additional Features

Feature	Description
Backup & Restore	Portable snapshot and restore of Kiro Crew state, for upgrades and machine migration
Knowledge Library	Semantic search over your own documents, folders, and generated artifacts
Web Deploy	Publish artifacts to a public HTTPS URL on your own AWS (private S3 + CloudFront + OAC)
Inbound Webhooks	Let an external system trigger an agent turn over HTTP — named tokens, HMAC request signing, a reversible off switch, ephemeral sessions, `register_hook` resume context
Feature Tips	Occasional personalized tips above the composer pointing at features you have not used yet
Follow-up Suggestions	Agent-proposed next steps above the composer: start in a new git worktree, add to this session, or skip
Queued-Message Editing	Edit, reorder, or cancel a chat message waiting in the queue before it runs
Cooperative Stop	Stop sends a cancel first and only hard-kills after a budget, so session state survives
Streaming Speech-to-Text	Live transcription partials in the dashboard input, with local Whisper or optional AWS Transcribe
Warm Pool	Keep kiro-cli processes pre-spawned so a new session starts instantly
Secrets Vault	Credentials encrypted on disk and refused to the agent, with a `secret://` reference left in `.env`

Settings reference

Settings is organized into tabs. Several have a full guide; the rest are documented by their own in-panel help.

Tab	Covers	Guide
Overview	Gateway status and the settings you change most	—
Imports	Bringing configuration in from another install	Snapshot and restore
Chat	Composer behaviour, queued messages, feature tips	Feature Tips
Display	Theme, language, and layout	—
Voice	Speech-to-text and spoken replies	Configuration
Notifications	Where a proactive message is delivered	—
Shortcuts	Keyboard bindings	—
Skills	Whether sessions auto-generate skills, and whether a generated one needs your approval; installed skills live under Agent Capabilities	Skills
Channels	Per-channel setup and access control	Channel capabilities
Browser	Installing the browser engine and the attach token	Browser control
Computer Use	Driving native desktop apps; off by default	Computer use
Webhooks	Inbound tokens and request signing — hidden unless you enable it under Feature Previews	Inbound webhooks
Instances	Additional gateways this dashboard can reach	—
Privacy	What leaves the host	Snapshot and restore
Security	The sandbox, denied commands, and the audit log	Blocked commands
Secrets	The encrypted credential vault	Secrets vault
Developer	The Developer Mode consent switch, plus an optional local-gateway toggle; turning it on adds a separate Developer page that holds logs, metrics, storage and the rest	Dashboard
Releases	Update channel and version	Getting Started
About	Version and links	—

Chat Channels

Besides the dashboard and CLI, Kiro Crew ships channel integrations for Slack, Discord, Telegram, Teams, Webex, WeCom, Weixin, iMessage, WhatsApp, and Feishu. They all share one channel-neutral core, so a capability a channel lacks degrades gracefully rather than failing the turn.

Guides

Getting Started: installation, first-time setup, running in the background
Configuration: config file reference, environment variables, sandbox
Use Cases: real-world workflows from the community
Troubleshooting: common issues and fixes
Blocked commands and credential access: why a command was refused, what the agent is told to do instead, and how to check your AWS or SSO credential setup
MCP Apps: render interactive MCP tool output (diagrams, viewers, forms) in chat, the two gates that enable it, what a server must declare, and why output stays plain text otherwise

Security

OS-level sandbox for the agent process, layered on top of kiro-cli's own
Credential redaction across every LLM output path
HMAC-SHA256 signed, IP-pinned dashboard tokens
Denied-command rules enforced at Kiro Crew's own PreToolUse gate, with audit logging, and a refusal that names the sanctioned path instead of only the rule (Blocked commands)
Prompt-injection credential-exfiltration protection
Slack access is owner-only: multi-user access and open channels are refused
An enabled app runs in-process with the gateway's own privileges, so enabling one is a trust decision. Third-party app execution is deny-by-default, and every app load is audited