← Coding

KiroKiro

AWS

编程Coding IDEIDE 多模型Multiple models 混合Hybrid partial lifecycle: active 核验Checked 2026-10-23

AWS 出的 agentic 工程环境：一套 harness 同时撑起 IDE / CLI / Web / Mobile 四个端。

两个招牌：

Spec 驱动 —— 先出 requirements / design / tasks 三段式，再写代码
Cloud Session —— 关掉笔记本，任务还在云上跑

Kiro is a desktop IDE — officially "a desktop development environment built on a VS Code foundation" — but it is framed as one agent, every surface: the IDE, CLI, Web and Mobile apps are all front ends to the same unified agent harness, connected over the open Agent Client Protocol (ACP). Its headline workflow is spec-driven development, turning a prompt into "requirements, architectural designs, and sequenced tasks", and its cloud sandbox sessions keep working "whether or not you're connected".

适合与不适合When to use it, when not to

在 AWS 生态里做正式工程交付、习惯「先出 spec 再实现」的团队。需要云端会话在关机后继续跑的人（Cloud sessions，最多 10 并发）。想把 steering 与权限规则随仓库分发的团队（.kiro/ 可提交）。

不适合想接自有 API Key 或跑本地模型的人。官方 models 与 pricing 两页都没有 BYOK 说明。也不适合要求额度口径透明的重度用户。credit 按模型倍数消耗，官方未给固定请求数换算。

Teams doing formal engineering delivery in the AWS ecosystem who want specs before implementation. People who need a cloud session that keeps running after they close the laptop (Cloud Sessions, up to 10 concurrent). Teams that want steering and permission rules shipped with the repository (.kiro/ is committable and consistent across surfaces). Free tier suits low-volume trials (50 credits/month, limited to open-weight models and Claude Sonnet 4.5).

Not a fit for anyone who needs their own API key or local models — neither the official models page nor the pricing page documents BYOK; nor for heavy users who need a transparent quota figure, since credits are consumed by model multiplier (up to 4.4x) with no fixed request-count equivalent.

中文生态China availability 国产 / 国内可直接用 / 备案状态 · 快照domestic / direct access / CAC filing · snapshot 2026-10-20

出品属地Origin: 国外出品foreign-made　国外主体（网信办公示名单未见对应备案主体）。foreign entity (no matching filing entity in the CAC registry).
国内使用Direct access: 大陆可直连direct OK　官网直连实测 200。official site measured 200 direct.
备案状态CAC filing: 未见于备案名单not in filing list：未见于网信办公示名单（2026-10-20 检索 13 批附件，https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm）；该产品未面向大陆公众提供服务，备案不适用。not in the CAC public registry (13 batch attachments searched 2026-10-20, https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm); the product does not serve the mainland public, so filing does not apply.

备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「大陆可直连」是本站从大陆网络直连实测的结果，不是厂商承诺。GitHub 类站点当日实测有波动，口径写在依据里。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise. GitHub-hosted sites fluctuated on the measurement day — the note says so.

固定坐标系Fixed coordinate system

8 个维度dimensions, 与同分区其他对象逐项可比each comparable item by item against others in the same section.

模型与开放条件Model & openness: 模型由 Amazon Bedrock 提供，可选多家，但要看你买哪档。

免费档只能用开放权重模型和 Claude Sonnet 4.5。付费档才有 premium 模型（含 Auto、Sonnet 5、Opus 5）。

官方 models 页列出的清单：

Auto —— 自动路由
OpenAI —— GPT-5.6（Sol / Terra / Luna）
Anthropic Claude —— Opus 5.5 / 5 / 4.8 / 4.7 / 4.6 / 4.5； Sonnet 5.5 / 5 / 4.6 / 4.5 / 4.0；Haiku 4.5

开放权重 —— DeepSeek 3.2、MiniMax M2.5 / M2.1、GLM-5、Qwen3 Coder Next

计费按 credit 倍数：Auto 是 1.0x 基准，GPT-5.6 Sol 最高 4.4x。

区域上，GPT-5.6 一律从美国服务。其余跟随你的 profile 区域（US / EU）。 Claude Fable 5.1 是 Enterprise 预览。

能不能自带 API Key（BYOK）：官方未说明。 （已查 models 与 pricing 两页。）
Models come through Amazon Bedrock; several vendors are selectable, but it depends on which tier you buy.

The free tier can only use open-weight models and Claude Sonnet 4.5. Premium models (including Auto, Sonnet 5, Opus 5) require a paid tier.

The list on the official models page:

Auto — automatic routing
OpenAI — GPT-5.6 (Sol / Terra / Luna)
Anthropic Claude — Opus 5.5 / 5 / 4.8 / 4.7 / 4.6 / 4.5; Sonnet 5.5 / 5 / 4.6 / 4.5 / 4.0; Haiku 4.5

Open-weight — DeepSeek 3.2, MiniMax M2.5 / M2.1, GLM-5, Qwen3 Coder Next

Billing is by credit multiplier: Auto is the 1.0x baseline, GPT-5.6 Sol tops out at 4.4x.

Regionally, GPT-5.6 is always served from the US. The rest follow your profile region (US / EU). Claude Fable 5.1 is an Enterprise preview.

Whether you can bring your own API key (BYOK): not stated officially.

(models and pricing pages checked.)
运行位置Where it runs: 独立桌面 IDE，官方原文明说「a desktop development environment built on a VS Code foundation」，不是装进别人编辑器的扩展。但官方把它定位为「one agent, every surface」： IDE / CLI / Web / Mobile 都是同一个 unified agent harness 的前端，本地客户端经 stdio 走 Agent Client Protocol（ACP），Web / Mobile 经 WebSocket 连云端沙箱。 IDE 与 CLI 默认在本机运行 harness。工具操作本地文件与 shell。两者也可创建并接入云端 sandbox 里的 Cloud Session。因为边界是标准 ACP。官方称 JetBrains IDE 与 Zed 也能把 Kiro 当 agent 用。
A standalone desktop IDE; the official wording explicitly says "a desktop development environment built on a VS Code foundation" — not an extension inside someone else's editor. But the official positioning is "one agent, every surface": IDE / CLI / Web / Mobile are all front ends of one unified agent harness; local clients go over stdio via the Agent Client Protocol (ACP), while Web / Mobile connect over WebSocket to cloud sandboxes. The IDE and CLI run the harness locally by default, with tools operating on local files and shell. Both can also create and attach Cloud Sessions inside cloud sandboxes. Because the boundary is standard ACP, the official pages say JetBrains IDEs and Zed can also use Kiro as their agent.
本地文件Local files: 它通过 IDE 访问你的工作区文件。

代码理解分两层。 第一层是内置的 Tree-sitter，覆盖 18 种语言，提供符号模糊搜索、文档大纲、AST 结构搜索与改写、代码库概览。这些能力跨 surface 都能用。第二层是 LSP（可选），能做查引用、跳定义、重命名、诊断、悬停文档 —— 但只在 IDE 与 CLI 里，而且要自己装语言服务器。

索引有专页说明：打开项目就自动索引源码、文档、配置与依赖；文件变更或外部改动会增量重索引。也可以手动 Force Re-Index / Rebuild。想让 agent 别碰某些文件，用 .kiroignore（gitignore 风格）排除。

索引会不会上传到云端：官方未说明。 （已查 codebase indexing 与 how-kiro-works 两页。）
It reaches your workspace files through the IDE.

Code understanding has two layers. The first is built-in Tree-sitter, covering 18 languages, providing fuzzy symbol search, document outlines, AST structural search and rewrite, and codebase overviews. These work across surfaces. The second layer is LSP (optional): find references, go to definition, rename, diagnostics, hover docs — but only in the IDE and CLI, and you install the language servers yourself.

Indexing has its own page: opening a project automatically indexes source, docs, configuration and dependencies; file changes or external edits trigger incremental re-indexing. Manual Force Re-Index / Rebuild is also available. To keep the agent off certain files, exclude them with .kiroignore (gitignore style).

Whether the index is uploaded to the cloud: not stated officially.

(codebase indexing and how-kiro-works pages checked.)
关机后的任务Tasks after shutdown: 关掉笔记本，任务还在跑 —— 这是它最特别的一点。

会话跑在厂商托管的云端 sandbox 里，属于账号而不是某个 app。所以 IDE / CLI / Web / Mobile 都能接入同一个会话，你合上电脑、改用手机或终端回来接着看都行。

官方给出的硬数字：

最多 10 个并发 Cloud Session
云端算力不额外收费（含在现有套餐里）
从 IDE 创建需 IDE v1.0.293+；从 CLI 需 CLI v2.17+

不过，跑在本机的那套 harness 仍然依赖你自己的机器，退出就停。

官方 Cloud sessions 页原文：「The agent keeps working in the sandbox whether or not you're connected」
Close the laptop and the task keeps running — that is its most distinctive point.

Sessions run in vendor-hosted cloud sandboxes and belong to the account, not to one app. So IDE / CLI / Web / Mobile can all attach to the same session; you can close the lid and pick it up from your phone or a terminal.

Hard numbers given officially:

Up to 10 concurrent Cloud Sessions
Cloud compute carries no extra charge (included in existing plans)
Creating from the IDE needs IDE v1.0.293+; from the CLI needs CLI v2.17+

However, the harness running locally still depends on your own machine and stops when you exit.

Official Cloud sessions page wording: "The agent keeps working in the sandbox whether or not you're connected"
工具与扩展Tools & extensions: 内置四类工具：文件、shell、web、code。

扩展机制成体系，一共六种：

MCP —— 本地 stdio 与远程 HTTP/SSE；支持 server 提供的 prompts / resources 与 elicitation；有 kiro:// 一键安装链接

Powers —— 打包了知识的 MCP server，按需加载
Skills —— 开放标准的可移植指令包
Custom agents 与 sub-agents
Specs
Hooks

Hook 的触发点官方列全了：Prompt Submit、Agent Stop、Session Start、 Session End（CLI V3）、Agent Spawn、Pre / Post Tool Use、 File Create / Save / Delete、Pre / Post Task Execution、Manual。

CLI 另有 headless 模式（--no-interactive，配 API key），可以放进 CI/CD。

⚠ Mobile 端不支持 MCP 与本地文件工具。
Four built-in tool classes: files, shell, web, code.

The extension mechanisms form a system, six in total:

MCP — local stdio and remote HTTP/SSE; supports server-provided prompts / resources and elicitation; kiro:// one-click install links

Powers — MCP servers packaged with knowledge, loaded on demand
Skills — open-standard portable instruction packs
Custom agents and sub-agents
Specs
Hooks

The Hook trigger points are fully listed officially: Prompt Submit, Agent Stop, Session Start, Session End (CLI V3), Agent Spawn, Pre / Post Tool Use, File Create / Save / Delete, Pre / Post Task Execution, Manual.

The CLI also has headless mode (--no-interactive, with an API key) for CI/CD.

⚠ The Mobile surface does not support MCP or local file tools.
上下文与记忆Context & memory: Specs 是它的招牌。 每个 spec 生成三份文件，走「需求 → 设计 → 任务」：

requirements.md（或 bugfix.md）
design.md
tasks.md —— 支持按依赖图分波并行执行

Steering 提供持久的项目上下文。它放在两个位置：工作区 .kiro/steering/、全局 ~/.kiro/steering/。内置 product / tech / structure 三份基础文件。支持 always / fileMatch / manual / auto 四种 inclusion 模式，并兼容 AGENTS.md。

其它上下文相关能力：subagents（各自独立上下文窗口）。另有 compaction。（自动摘要旧历史，单向不可回滚）· checkpoints / rewind · /sessions 做会话管理。CLI 里另有实验性的 knowledge 库（语义检索）。

上下文窗口的具体数值，官方 docs 没有单列。 （已查 compaction 与 models 两页。）
Specs are its signature. Each spec generates three files, following "requirements → design → tasks":

requirements.md (or bugfix.md)
design.md
tasks.md — supports wave-parallel execution along the dependency graph

Steering provides persistent project context. It lives in two places: workspace .kiro/steering/, global ~/.kiro/steering/. Three base files ship built in: product / tech / structure. Four inclusion modes are supported — always / fileMatch / manual / auto — and AGENTS.md is compatible.

Other context-related capabilities: subagents (each with its own context window). There is also compaction (auto-summarizing older history, one-way and not reversible) · checkpoints / rewind · /sessions for session management. The CLI additionally has an experimental knowledge base (semantic retrieval).

Concrete context-window numbers are not broken out in the official docs. (compaction and models pages checked.)
权限与限制Permissions & limits: 它是能力制权限，不是「信任 / 不信任」两档。 规则写 YAML，按 capability 配 match / exclude，effect 取 deny / ask / allow，优先级 deny > ask > allow。

官方列出的 capability：

文件：fs_read / fs_write
执行：shell / sandbox_network
联网：web_fetch / web_search
扩展：mcp / subagent / skill / power
其它：context / diagnostics

作用域共六个：Kiro 硬编码、administration、user、workspace、agent、session。

IDE 另有 Agent Autonomy 的 Autopilot / Supervised 两档。 工作区默认不被信任。信任之前不加载仓库内的 agents、steering、MCP、Skills。 headless 模式下，每个 ask 都按 deny 处理。

官方 permissions 页原文：capability-based，"replacing older binary trust models"
Permissions are capability-based, not a two-state trust/untrust model. Rules are written in YAML, configured per capability with match / exclude, effect taking deny / ask / allow, with precedence deny > ask > allow.

Capabilities listed officially:

Files: fs_read / fs_write
Execution: shell / sandbox_network
Networking: web_fetch / web_search
Extensions: mcp / subagent / skill / power
Other: context / diagnostics

Six scopes in total: Kiro hardcoded, administration, user, workspace, agent, session.

The IDE additionally has Agent Autonomy with Autopilot / Supervised tiers. Workspaces are untrusted by default. Before trust, repo-local agents, steering, MCP and Skills are not loaded. In headless mode, every ask is treated as deny.

Official permissions page wording: capability-based, "replacing older binary trust models"
适合什么任务Best for: 适合三类人：

在 AWS 生态里做正式工程交付、要「先出 spec 再实现」的团队
需要云端会话在关机后继续跑的人（Cloud sessions，最多 10 并发）
想把 steering 与权限规则随仓库分发的团队（.kiro/ 可提交，跨 surface 一致）

想低量试用的话，Free 档够用。每月 50 credits，限开放权重模型与 Sonnet 4.5。

不适合两类人。一是想接自有 API Key 或跑本地模型的。官方 models 与 pricing 两页都没有 BYOK 说明。二是要求额度口径透明的重度用户。credit 按模型倍数消耗，官方没给「固定请求数」的换算。
Fits three kinds of people:

Teams doing formal engineering delivery in the AWS ecosystem who want "spec first, implementation second"
People who need cloud sessions to keep running after shutdown (Cloud sessions, up to 10 concurrent)
Teams that want steering and permission rules distributed with the repo (.kiro/ is committable, consistent across surfaces)

For a low-volume trial, the Free tier suffices: 50 credits per month, limited to open-weight models and Sonnet 4.5.

Does not fit two kinds of people. First, those wanting to plug in their own API keys or run local models — neither the official models nor pricing pages has any BYOK statement. Second, heavy users who need transparent quota accounting — credits are consumed at model multipliers, and no "fixed request count" conversion is given officially.

头号误解Top misconception

把 Kiro 当成纯 IDE 工具，它另有 CLI / Web / Mobile 与 Crew，Cloud Session 还能关机后续跑Treating Kiro as an IDE-only tool: there is also CLI / Web / Mobile and Crew, and a Cloud Session keeps running after shutdown
以为 credits 按模型「包干」，官方是统一 credit、各模型按倍数（最高 4.4x）消耗Assuming credits are model-specific — Kiro uses one credit pool and each model consumes it by a multiplier (up to 4.4x)
把 Kiro Free 当长期可用的免费档，它每月仅 50 credits 且限特定模型Taking Kiro Free as a long-term free tier — it is 50 credits/month, limited to open-weight models and specific Claude models

未知项清单Known unknowns

是否支持自有 API Key（BYOK）与本地模型——官方未说明（已查 /docs/models 与 /pricing 两页）
IDE 代码索引是否上传云端——官网未公开索引位置（已查 codebase indexing 与 how-kiro-works 两页）
各模型上下文自动压缩的触发阈值（官方未给数字）
Crew 的完整能力边界（只核到首页入口与 changelog 条目）

Whether own API keys (BYOK) and local models are supported — not stated officially (/docs/models and /pricing checked)
Whether IDE code indexing is uploaded to the cloud — the official site does not disclose index location (codebase indexing and how-kiro-works checked)
Automatic compaction trigger thresholds per model (no numbers given officially)
The full capability boundary of Crew (the 2026-10-08 check only captured the homepage entry and changelog entries)

证据来源Evidence sources

判断可回到以下一手源复核。本站核验日Every judgement can be rechecked against the primary sources below. Verified on 2026-10-23，内容更新日content updated on 2026-10-08。

类型Type	名称Name	链接Link
changelog	Kiro · Changelog（IDE 1.2 / CLI 2.27 / Workflows / Opus 5.5，2026-10-08 取到正文）Kiro · Changelog (IDE 1.2 / CLI 2.27 / Workflows / Opus 5.5, body captured 2026-10-08)	https://kiro.dev/changelog/
pricing	Kiro · 官方定价页（Free / Pro / Pro+ / Pro Max / Power 五档 + credit 说明，2026-10-08 核）Kiro · official pricing page (Free / Pro / Pro+ / Pro Max / Power five tiers + credit notes, verified 2026-10-08)	https://kiro.dev/pricing/
docs	Kiro · 官方文档首页Kiro · official docs home	https://kiro.dev/docs/
docs	Kiro · How Kiro works（unified harness、ACP、各 surface 运行位置）Kiro · How Kiro works (unified harness, ACP, where each surface runs)	https://kiro.dev/docs/how-kiro-works/
docs	Kiro · Specs（requirements / design / tasks 三段式与并行任务）Kiro · Specs (requirements / design / tasks three-part structure and parallel tasks)	https://kiro.dev/docs/specs/
docs	Kiro · Steering（工作区 / 全局 steering、四种 inclusion 模式、AGENTS.md）Kiro · Steering (workspace / global steering, four inclusion modes, AGENTS.md)	https://kiro.dev/docs/steering/
docs	Kiro · Hook triggers（Prompt Submit / Pre Tool Use / File Save 等完整触发点）Kiro · Hook triggers (Prompt Submit / Pre Tool Use / File Save and the full trigger set)	https://kiro.dev/docs/hooks/types/
docs	Kiro · MCP（stdio / HTTP / SSE、`kiro://` 安装链接）Kiro · MCP (stdio / HTTP / SSE, `kiro://` install links)	https://kiro.dev/docs/mcp/
docs	Kiro · Models（多厂商模型清单与 credit 倍数）Kiro · Models (multi-vendor model list and credit multipliers)	https://kiro.dev/docs/models/
docs	Kiro · Permissions（capability-based 规则与六作用域）Kiro · Permissions (capability-based rules and six scopes)	https://kiro.dev/docs/permissions/
docs	Kiro · Cloud sessions（云端 sandbox、10 并发、跨 surface）Kiro · Cloud sessions (cloud sandbox, 10 concurrent, cross-surface)	https://kiro.dev/docs/cloud-sessions/
docs	Kiro · Code intelligence（Tree-sitter / LSP 两层）Kiro · Code intelligence (Tree-sitter / LSP two layers)	https://kiro.dev/docs/tools/code-intelligence/

实测记录Measured runs

本档未跑端到端编码任务；能力结论来自官方文档，编码向测量协议见 protocols/coding/_protocol.md。

实测建议：重点测两点。

spec 三段式的实际产出质量
Cloud Session 从 IDE 创建后，关掉本机、再从手机接回时的状态完整性

这两点是官方最明确的承诺。

This site has not completed measured runs. The test protocol is at protocols/coding/_protocol.md.

Suggested measurement: focus on two things.

The actual output quality of the spec three-part structure
State integrity when a Cloud Session created from the IDE is picked back up from a phone after closing the local machine

These two are the strongest promises the official pages make.

四个 surface，一个 harnessFour surfaces, one harness

官方把形态讲得很明确——「one agent, available everywhere you work」：

surface	官方定位
IDE	基于 VS Code 的独立桌面 IDE（built on a VS Code foundation）
CLI	终端 TUI，另有 headless 模式进 CI/CD
Web	浏览器里的云端 agent，跑在托管 sandbox
Mobile	手机上启动与 steer 会话

关键设计：IDE / CLI 默认在本机跑 harness，Web / Mobile 在云端 sandbox 跑。 但四者是同一个 harness 的前端，靠开放协议 ACP 连接。因此官方称 JetBrains IDE 与 Zed 也能把 Kiro 当 agent 用。

The official pages state the form clearly — "one agent, available everywhere you work":

surface	official positioning
IDE	a standalone desktop IDE based on VS Code (built on a VS Code foundation)
CLI	a terminal TUI, plus headless mode for CI/CD
Web	a cloud agent in the browser, running in a hosted sandbox
Mobile	start and steer sessions from your phone

Key design: the IDE / CLI run the harness locally by default; Web / Mobile run in cloud sandboxes. But all four are front ends of the same harness, connected by the open protocol ACP. Hence the official claim that JetBrains IDEs and Zed can also use Kiro as their agent.

招牌：spec-driven developmentThe signature: spec-driven development

官方首页原话：把 prompt 变成「requirements, architectural designs, and sequenced tasks」。每个 spec 落地为三份文件：

requirements.md（或 bugfix.md）——用户故事与验收标准
design.md——技术架构、时序图与实现考量
tasks.md——离散可追踪任务，支持按依赖图分波并行执行

配套还有 steering 文件（.kiro/steering/ 的持久项目上下文，四种 inclusion 模式）与 agent hooks（十余个触发点，从 File Save 到 Pre Task Execution）。

The official homepage wording: turn a prompt into "requirements, architectural designs, and sequenced tasks". Each spec lands as three files:

requirements.md (or bugfix.md) — user stories and acceptance criteria
design.md — technical architecture, sequence diagrams and implementation considerations
tasks.md — discrete trackable tasks, supporting wave-parallel execution along the dependency graph

The supporting cast includes steering files (persistent project context in .kiro/steering/, four inclusion modes) and agent hooks (a dozen-plus trigger points, from File Save to Pre Task Execution).

核验说明Verification notes

本条目 confidence: partial 的原因：

✅ 已核验：五档档位结构与 credit 规则、模型清单与倍数、specs / steering / hooks 结构、 MCP 传输方式、capability-based 权限、Cloud sessions 并发数字、「built on a VS Code foundation」的形态表述

❌ 未取到：BYOK / 自有 Key 是否支持、IDE 索引是否上云、压缩触发阈值、Crew 的完整能力面

核验页：已核到的官方页：kiro.dev/pricing、/changelog、/docs、/docs/how-kiro-works、 /docs/specs、/docs/steering、/docs/hooks/types、/docs/mcp、/docs/models、 /docs/permissions、/docs/cloud-sessions、/docs/tools/code-intelligence。

Why this entry is confidence: partial:

✅ Verified: the five-tier structure and credit rules, the model list and multipliers, the specs / steering / hooks structure, MCP transport types, capability-based permissions, Cloud sessions concurrency numbers, the "built on a VS Code foundation" form statement

❌ Not obtained: whether BYOK / own keys are supported, whether IDE indexing goes to the cloud, compaction trigger thresholds, the full Crew capability surface

Verified pages: official pages checked the 2026-10-08 check: kiro.dev/pricing, /changelog, /docs, /docs/how-kiro-works, /docs/specs, /docs/steering, /docs/hooks/types, /docs/mcp, /docs/models, /docs/permissions, /docs/cloud-sessions, /docs/tools/code-intelligence.