What's New in Kiro

Kiro ships something most weeks, across five surfaces and a model list. Kiro’s own changelog records every release in detail. This page is the short version: what changed, whether it affects you, and what to do. Newest first, covering the last month. Older milestones are in History.

Needs action

These changes can break an existing setup or close a security gap. Check them first.

Kiro CLI 3.0 (end of October): V3 becomes the default. Kiro announced on 9 October that V3, the terminal UI built on the same agent harness as the IDE and Web, becomes the default with CLI 3.0 by the end of October. Since 12 October a startup prompt has been inviting people to switch early (2.30.0). CLI 3.0 also retires inline shell autocomplete, the CLI desktop app, and Classic sessions. Choosing “Don’t ask again” does not keep you on V2; your setup is still upgraded when 3.0 arrives. Affects you if you use custom agents, --classic in an alias, the use_aws tool, or an ACP integration. Do this: run kiro-cli --v3 on a real task now. Upgrade agents with /upgrade-agent (or kiro-cli agent upgrade <dir> for a folder, with --dry-run first), then run /upgrade-agent diagnostics for permission warnings. V3 has no use_aws tool, so recreate those rules as shell permission rules. ACP clients start the V3 server with kiro-cli acp --agent-engine=v3. See the CLI Guide.

Claude Sonnet 4.0 end of life (14 October). Sonnet 4.0 is gone from Kiro. The picker marks retired models [EOL]. Affects you if a custom agent, a default model setting, or a workflow recipe names Sonnet 4.0. Do this: point it at Sonnet 5.5, Sonnet 5, or Auto. See Choosing a Model.

Kiro Crew 0.8.0 (27 September): two changed defaults. Owner-written autoApprove lists in mcp.json or an agent file now take effect, so those tools skip both the approval card and Crew’s own tool gate. Crew logs (a redacted event log per chat) are now on by default. 0.8.0 also fixed approved calls skipping a final denied-command and protected-path check. Affects you if you run Kiro Crew and have autoApprove entries you had forgotten about. Do this: review every autoApprove list before upgrading, or set mcp.honour_auto_approve to false and restart the Gateway to keep those tools behind the gate. Set KIROCREW_CREW_LOG=0 before starting the Gateway if you do not want crew logs.

Kiro CLI 2.24.0 (23 September): .env files no longer load automatically. Project .env files used to reach chat sessions, MCP servers, and tools on their own. Now Kiro starts with whatever environment your terminal gives it. Affects you if an MCP server or tool relied on a variable from a project .env. Do this: export the variables in your shell before running kiro-cli, and reference the exported values in the MCP server configuration.

Kiro Crew 0.7.2 (28 September): WhatsApp group fix. A configured WhatsApp group could previously let any member drive the agent. Affects you if you connected Crew to WhatsApp. Do this: upgrade Crew to 0.7.2 or later.

October 2026

Kiro IDE 1.4 (19 and 20 October): Memory. Kiro can now carry useful context, such as repository conventions, corrections, and approaches already ruled out, from one session to the next. Open the Memories icon in the activity bar to see what Kiro remembers. Since 1.4.10 the Kiro Agent: Memory setting defaults to Off; the other levels are Use and learn, Use but don’t learn, and Use only. 1.4.5 also shows a proposed workflow’s steps in a permission card before they run. Do this: turn Memory on if you are tired of restating the same facts, and keep rules that must always apply in steering. See Memory.

Kiro Crew 0.9.0 (16 October). Crewmate cards that show a goal and its progress without spending a chat turn, a per-crewmate Profile, previews of dynamic dashboards and remote crews as live windows, and a composer that warns before a prompt outgrows the model. New installs answer more briefly, and the Ultra verbosity level is retired. Do this: if you used Ultra, pick another Response Verbosity level. Some settings moved: Agent Capabilities is now Customize, and Warm Pool is now Pre-started Sessions under Agent Harness.

Kimi K3 (15 October). Moonshot AI’s open-weight model: 1M context, text and image input, 2.4x, experimental, Pro and above, US geography only. Do this: nothing required. It costs more than any Opus, so it is not a budget pick despite being open-weight.

Kiro IDE 1.3.11 (15 October): Markdown Review. Open any workspace .md file in a rendered panel, comment on exact passages, and send the comments to Kiro in one batch. It works on spec documents from the Review button. Generate Commit Message also works before you stage anything. Do this: try it on your next requirements.md or design.md instead of pasting feedback into chat.

Kiro CLI 2.32.0 (15 October). Allow a whole folder when a V3 session asks to write a file, run cloud sessions from non-interactive commands (--cloud, --repo), and preview a workflow’s plan before any step runs. Workflow review loops now pause after two rejected reviews. Do this: nothing required.

Kiro CLI 2.31.0 and IDE 1.3 (13 October): local command sandbox. Agent-run shell commands and their child processes can run inside an operating-system sandbox, with outbound network blocked, limited to common development hosts, or open. The agent itself, built-in file tools, and remote MCP servers stay outside it, and Kiro says plainly that it is not complete isolation. The CLI also gained /autonomous on, which turns the sandbox on and stops asking you to approve each shell command, plus KIRO_HOME for separate profiles. The IDE gained /workflow-run. Do this: turn the sandbox on before you let an agent run unattended. See Turn on the command sandbox.

Kiro Web (12 October). Choose a fallback model per model, change the model of a single workflow step, and recover a loop at its iteration cap with Run N more or Stop looping and continue. Do this: set a fallback for the model you use most, so an outage does not stall a long session.

FedRAMP Class C (12 October). Kiro is in scope for FedRAMP Class C in US East (N. Virginia). Do this: nothing, unless your organisation needs that authorisation to adopt Kiro.

Kiro CLI 2.29.0 (9 October): Memory in V3. /memories → config chooses read & write, read only, or off; list shows what is stored. Admins can now turn off device-flow sign-in per profile. Do this: nothing required. In an untrusted workspace Kiro can read memories but not change them.

Opus 5.5 and Sonnet 5.5 in AWS GovCloud (US) (9 October), in the IDE and CLI.

Kiro IDE 1.2.56 (8 October) and CLI 2.28.0 (5 October): hook categories. Hooks that run before or after a tool can match a category such as shell, read, or write, or an MCP server, instead of listing every tool ID. CLI 2.28.0 also made V3 chat always delegate through Workflows once they are enabled. Do this: if you keep long tool-ID lists in a hook matcher, replace them with a category. See Hooks.

GPT-5.6 Luna back to 0.6x. By 21 October Kiro’s model documentation listed Luna at 0.6x (1.2x above 272K tokens), down from 1.1x in September. No changelog entry announced it. Do this: if you dropped Luna after the September re-pricing, it is cheaper than Auto again. The credit calculator uses the new number.

Claude Sonnet 5.5 (2 October). 1M context at 1.3x, the same price as every Sonnet, with output more than 30% faster than Sonnet 5.

Kiro CLI 2.27.0 (1 October). Saved prompts became slash commands, and V3 steering gained line-range #[[file:...]] references and #[[folder:...]] listings.

Late September 2026

Workflows in the IDE, CLI, and Web (30 September). Reusable multi-step plans in which each step runs as its own agent session in the background. Off by default everywhere. IDE 1.2 also asks before the agent edits Kiro’s own configuration, and untrusted workspaces do not load workflow recipes. See Workflows.

Claude Opus 5.5 (28 September). 1M context at 2.0x, the first Opus priced below 2.2x.

Kiro CLI 2.25.0 (28 September). /powers install and uninstall, V3 output styles, and a SessionEnd hook trigger.

Kiro IDE 1.1.70 and Crew 0.7.0 (24 September). A default timeout for agent-run terminal commands in the IDE. In Crew, settings apply without a restart, queued work survives Gateway restarts, and skills.max_triggered now defaults to 0.

Kiro Web (22 September). A reasoning effort selector for supported models. See set the reasoning effort.

How this page stays current

The site re-reads Kiro’s IDE, CLI, Web, Crew, and model changelogs about every two weeks, records sources in its research notes, and updates this page and the affected chapters. The site changelog lists each pass. If something here disagrees with kiro.dev/changelog, trust Kiro’s page.