Kiro Release Notes

Added support for GLM-5, a sparse mixture-of-experts model with a 200K context window designed for complex systems engineering and long-horizon agentic tasks. GLM-5 excels at processing repository-scale context and maintaining coherence during multi-step tool use across large codebases, making it a strong choice for cross-file migrations, full-stack feature development, and legacy refactoring where the model needs to hold the full picture. Available with experimental support in both the Kiro IDE and Kiro CLI. Inference runs in us-east-1 (N. Virginia) with a 0.5x credit multiplier. Restart your IDE or CLI to access it from the model selector.

This release brings an experimental refreshed terminal UI behind the --tui flag. The new experience features a live status bar, rich markdown rendering with syntax-highlighted code blocks, interactive panels for managing context and sessions, and contextual overlays for tools and help. You also get /chat new for starting fresh conversations without restarting the CLI, and --list-models for quickly checking which models are available.

New Terminal Experience

Get a richer chat experience with the new TUI. Agent responses render with full markdown, syntax-highlighted code blocks, tables, and nested formatting. Tool calls show descriptive titles and progress indicators as they run. Interactive panels let you browse context, switch sessions, and manage tools without leaving the terminal. The classic interface remains the default, so you can switch back at any time. Try it with kiro-cli --tui or set it permanently with kiro-cli settings chat.ui "tui".

Learn more ->

This release gives enterprise teams centralized control over MCP server access and model availability, and adds document attachment support to chat.

MCP Registry Governance

Enterprise administrators using IAM Identity Center can now control which MCP servers their organization is allowed to use. Create a JSON registry file listing approved servers, host it over HTTPS, and configure the URL in the Kiro console. The registry supports remote (HTTP) and local (stdio) servers across npm, PyPI, and OCI package types, with ${VAR} placeholders for user-specific values like auth tokens. Kiro syncs with the registry every 24 hours, enforces version-pinned access, and works alongside the existing MCP on/off toggle to give you full control at the org or account level.

Model Governance

Enterprise administrators can now control which AI models are available to users in their organization. Toggle on model access management in the Kiro console under Settings > Shared settings > Model availability, then curate an approved list of models. You can also set a default model that auto-applies across all clients. This is especially relevant for data residency requirements — experimental models using global cross-region inference can be removed from the approved list until they move to GA with regional inference. Once enabled, only approved models appear in the model selector across both the IDE and CLI.

Document Attachments

You can now attach documents directly to chat messages by pasting or dragging files into the input. Supported formats include PDF, CSV, DOC, DOCX, XLS, XLSX, HTML, TXT, and Markdown. Documents are sent to the model as native document blocks, so the agent can read and reason over their contents. You can attach up to 5 documents per message and mix them with text and images in the same prompt.

This release streamlines how you create and manage agents. Describe what you need and Kiro CLI generates your agent config, use the new session settings tool to tweak preferences mid-conversation without editing config files, and control exactly which shell commands you trust with granular scoped permissions.

Simplified Agent Creation

The /agent create command now defaults to AI-assisted mode, merging the previous /agent generate workflow into a single command. Describe what you want your agent to do and Kiro generates the agent configuration for you. Pass --manual to use the previous editor-based creation flow. You can also specify creation arguments directly at invocation time to bypass the interactive menu entirely.
Learn more ->

Granular Tool Trust

When Kiro asks to use a tool, you now get an interactive picker to choose how broadly to trust it. For shell commands, select from tiered scopes -1 trust the exact command, the command with any arguments, or the base command with wildcards. For read and write tools, scope trust to specific file paths, the containing directory, or the entire tool. The picker adapts to each action, skipping tiers that aren't meaningful, and handles chained shell commands automatically.
Learn more ->

Session Settings Tool

You can now ask the agent to adjust settings temporarily within your current session -1 change model preferences, toggle features, or tweak behavior without modifying your config files. All session overrides apply in-memory and reset automatically when the session ends.
Learn more ->

This release introduces new spec workflows that let you design features architecture-first and fix bugs with structured root cause analysis. Supervised mode now lets you review changes at the hunk level for precise control. New hook triggers automate work around spec tasks, and MCP servers can now surface prompts, templates, and elicitation support directly in chat.

Design-First Feature Specs

A new Design-First workflow lets you start a feature spec from technical architecture instead of requirements. Provide a high-level or low-level design, pseudocode, or system diagram and Kiro derives feasible requirements from it. This is ideal when you have strict non-functional constraints, an existing design to port, or want to think through the architecture before defining user-facing behavior.
Learn more ->

Bugfix Specs

Specs now include a dedicated Bugfix workflow. Describe the issue you're seeing and Kiro walks you through root cause analysis, fix design, and regression prevention. The result is a bugfix.md that captures current behavior, expected behavior, and what must remain unchanged, giving the agent clear guardrails before it writes a single line of code.
Learn more ->

Hunk-Based Review in Supervised Mode

Supervised mode now presents file changes as individual hunks instead of full-file diffs. Each hunk is a logical group of related lines that you can independently accept, reject, or discuss with inline chat. You can also accept changes at the file level or accept all changes at once. This gives you precise control over exactly which parts of a change to keep.
Learn more ->

Pre and Post Task Execution Hooks

Two new hook triggers let you automate work around spec task execution.
Pre Task Execution fires before a task begins, so you can run setup scripts or validate prerequisites.
Post Task Execution fires after a task completes, letting you run tests, linting, or notify external systems automatically. Combine these with the existing hook actions to build end-to-end automation around your spec workflows.
Learn more ->

MCP Prompts, Resource Templates, and Elicitation

MCP servers can now provide prompts and resource templates that appear in the context provider (#) menu in chat. Select a prompt to insert pre-built instructions, or fill in a resource template's parameters to pull in specific content as context. During tool execution, servers can also request additional input from you through elicitation. Kiro supports elicitation so servers can request the information they need without interrupting your workflow.
Learn more ->

Kiro is now available in AWS GovCloud (US-East) and AWS GovCloud (US-West) regions, enabling government agencies and contractors to use Kiro within their compliance boundaries.

GovCloud Authentication

GovCloud Authentication
Authenticate using AWS IAM Identity Center with a GovCloud Start URL (containing us-gov-home). Kiro uses the same installer for both commercial and GovCloud regions—IAM Identity Center authentication automatically routes traffic to the appropriate region. Individual login methods like GitHub, Google, and AWS Builder ID are not available in GovCloud regions.
Learn more ->

Inference Routing

Inference Routing
For GovCloud customers, inference requests are processed using Amazon Bedrock in AWS GovCloud (US-West). Your content remains stored in the region where your Kiro profile was created, with all cross- region communication encrypted via TLS 1.2+.
Learn more ->