Changelog

Claude Sonnet 4 has reached end of life in Kiro as of October 14, 2026. If your work depends on it, choose another available model before continuing. Claude Sonnet 5.5 is the latest Sonnet model. See Model lifecycle for lifecycle stages and end-of-life dates.

This release lets you limit what agent-run shell commands can reach and start saved Workflows from chat. It also shows live session status on editor tabs, explains model fallbacks, loads MCP tools on demand, and fixes Workflow, chat, and account issues.

Turn on the local command sandbox to limit what agent-run shell commands and their child processes can access. Choose whether outbound network access is blocked, limited to common development hosts, or open. You can also run local stdio MCP servers inside the same sandbox.

The agent, built-in file tools, and remote MCP servers stay outside the sandbox. Permission rules still govern agent actions. The sandbox reduces risk but cannot guarantee complete isolation, so keep reviewing proposed commands. Select Command sandbox in the status bar to open the settings and confirm the sandbox is active. Learn more ->

Type /workflow-run, choose a recipe, complete its inputs, and start the run. Kiro opens Workflows on the new run. Learn more ->

This release makes the local command sandbox and local autonomous mode available to every V3 user. It also loads MCP tools on demand, extends KIRO_HOME across V3 configuration and state, and fixes image, service, MCP, cloud-session, and sign-in issues.

Run /sandbox to place agent shell commands and their child processes inside an operating-system sandbox. Choose whether outbound network access is blocked, limited to common development hosts, or open. You can also run local stdio MCP servers inside the same sandbox.

The agent, built-in file tools, and remote MCP servers stay outside the sandbox. Permission rules still govern agent actions. The sandbox reduces risk but cannot guarantee complete isolation, so keep reviewing proposed commands. Learn more ->

Run /autonomous on in a local V3 session when you want Kiro to work through a task without asking you to approve each shell command. Kiro switches to the Default agent and turns on the local command sandbox first. Learn more ->

Kiro Web now lets you choose a fallback model and change the model a workflow step runs on. Reopened sessions show their conversation while the sandbox starts.

You can also recover a workflow loop that hit its iteration cap, and sessions you keep using stay available.

Open Settings > Model, choose a model from the list, and pick its fallback. If the selected model is unavailable, the fallback answers only that turn, and Kiro returns to the selected model on the next one. If the selected model refuses a request, the conversation stays on the fallback for the rest of the session. You can remove a fallback at any time. Learn more ->

With Workflows turned on, the selected step's composer now has a model picker, including effort and Thinking when the model supports them. Stop the step before changing its model, then send a message to continue. The choice applies only to that step. Learn more ->

When you reopen a session whose sandbox has stopped, Kiro Web shows a preview of the saved conversation while the sandbox starts. Once the sandbox is ready, the complete history replaces the preview. Learn more ->

When a repeat in a workflow pauses at its iteration cap, you now have two ways to continue. Run N more accepts a whole number from 1 to 1,000 and grants that many extra iterations, and Stop looping and continue ends the repeat and moves on to the next step. Learn more ->

This release shows a startup prompt inviting you to switch to V3 as part of a gradual rollout. It also adds V3 model fallback notices, keeps more of your V2 agent access when you upgrade agents to V3, and fixes screen reader, tmux, and knowledge search issues.

The startup prompt is rolling out gradually before V3 becomes the default with CLI 3.0 at the end of October, so you may not see it right away. When it appears, choose Switch to V3 and upgrade my configs to save V3 as your default harness, enable automatic agent upgrades, and start migrating your custom agent configurations. You can also try V3 anytime with kiro-cli --v3. Read the Kiro CLI V3 rollout announcement for details.

In V3 sessions, when a model refuses a request or is unavailable and the turn moves to another model, the transcript now shows a notice. You can tell that a different model continued the turn instead of wondering why a reply changed. Learn about model fallback notices.

Kiro is now in scope for FedRAMP Class C in the US East (N. Virginia) Region. US government agencies, contractors, and other organizations whose workloads are subject to FedRAMP Class C requirements can now use Kiro for AI-assisted software development.

FedRAMP is a US government-wide program that delivers a standard approach to security assessment, certification, and continuous monitoring for cloud products and services. See AWS services in scope by compliance program for current Kiro scope details, AWS compliance resources for program details, and FedRAMP on AWS for background on the FedRAMP program itself.

Claude Opus 5.5, Anthropic's most capable Opus model yet, is now available in the Kiro IDE and CLI across AWS GovCloud (US-East) and AWS GovCloud (US-West). It improves on Opus 5 for long-running agentic coding work and thinks adaptively on every request, deciding how much effort each task needs. It includes a 1M-token context window with a 2.0x credit multiplier, down from Opus 5's 2.2x. Update to the latest IDE or CLI version, then restart it to access the model from the model selector. Learn more ->

Claude Sonnet 5.5 is now available in the Kiro IDE and CLI across AWS GovCloud (US-East) and AWS GovCloud (US-West). It is Anthropic's fastest Sonnet model to date and a natural upgrade for teams already using Sonnet 5: output is more than 30% faster, real-world coding and knowledge work approaches Opus 5.5 performance, and writing is clearer. It is available with a 1M-token context window and a 1.3x credit multiplier. Update to the latest IDE or CLI version, then restart it to access the model from the model selector. Learn more ->

Kiro CLI 2.29.0 introduces Memory for local V3 sessions, so Kiro can carry useful context from your work, such as a repository's commands and conventions, into later chat sessions instead of asking you to restate it.

Kiro can now learn useful context from your work and recall it when a later session needs it. For example, if Kiro learns while debugging that a repository's integration tests fail until the local database is seeded, a later session can seed it first instead of rediscovering the cause. When Kiro uses a memory, the chat shows a Memory entry with command=get and the memory's title while Kiro is still working.

In a local V3 session, type /memories and select config to choose read & write, read only, or off. The read only option says "Use existing memories. No changes or automatic updates." off stops Kiro from using or changing memories. Access changes apply in your next session. With read & write, Automatic memory updates lets Kiro create and refine memories, including after a session ends; changes to that setting apply immediately. In a workspace you have not trusted, Kiro can read memories but cannot add, change, or delete them.

Select list to search, preview, and open stored memories. The list is read-only; to fix or delete a memory, ask Kiro in the chat.

Learn more ->

This release adds a feedback link for Workflows, fixes several chat, terminal, and source control issues, and improves workflow reliability, spec-only commands, and Hook matching.