让 AI 替我写代码Get AI to write code for me 23
编辑器里、终端里、或厂商云上替你干活的编码 agent。Coding agents that work in your editor, your terminal, or the vendor's cloud.
只看这一类 →Show only these →AI 工具AI Tools · Tools
现成的 AI 工具Ready-made AI tools
自己能跑的成品 AI agent，共用八维（模型接入 / 运行位置 / 本地文件 / 关机后任务 / 工具扩展 / 上下文 / 权限 / 适合场景）。按形态分三列——IDE 形态、CLI 形态、厂商云形态。前两者是同一产品的不同装法、不是竞争关系；厂商云形态则是另一个方向：不需要自己装，工作在别人的机器上做。Finished AI agents you can run yourself, compared on eight shared dimensions (model access / where it runs / local files / tasks after shutdown / tools and extensions / context / permissions / best fit). Grouped into three forms — IDE, CLI and vendor cloud. The first two are different install forms of the same product, not competitors; vendor cloud is a different direction entirely: nothing to install, the work happens on someone else's machine.
- 这里收录能接任务的产品。它要么在编辑器或终端里替你写代码，要么能被你装进自己的程序，要么能连上你的文件和账号。This site lists products that take on tasks. They either write code for you in your editor or terminal, or can be installed into your own program, or connect to your files and accounts.
- 纯聊天工具不走八维坐标系。它们没有「安装形态 / 工具清单」这些维度，硬套只会产出一堆「不适用」——本站给它们的是本页下方的三要素轻档案；更全的对话产品目录在 AI 导航。Plain chat products are not on the 8-axis grid. They lack dimensions like install form or tool inventory, and forcing them would produce a wall of "not applicable" — this site covers them with the three-essentials light entries below; a fuller directory of chat products lives in the AI Directory.
- 每条档案都是核验快照。字段逐条挂官方出处与核验日期，本站不实测、不排名，未知就写未知。Every entry is a verification snapshot. Each field carries an official source and a checked date; we run no benchmarks, publish no rankings, and write "unknown" when it is unknown.
按任务找Find by task
这一页有 27 个对象。下面的数字是本页真正收录的条数——点进去就是筛选后的结果，不是跨站的总数。This page holds 27 objects. The numbers below are what this page actually carries — clicking filters right here, not a site-wide total.
无人值守跑长任务Run long tasks unattended 4
关掉笔记本它还在跑的：厂商云形态与通用 harness。Keeps running after you shut your laptop: vendor-cloud forms and general harnesses.
只看这一类 →Show only these →对话问答Chat products
这些产品不走八维坐标系——它们没有「安装形态 / 工具清单」这些维度，硬套只会产出一堆「不适用」。三要素是它们真正可比的地方；价格常变，本站不存报价，一律看官方定价页。These products are not on the 8-axis grid — they lack dimensions like install form or tool inventory, and forcing them would produce a wall of "not applicable". The three essentials are where they are genuinely comparable; prices change often, so this site stores no quotes and always points to the official pricing page.
ChatGPTChatGPT
OpenAI · 轻档案 · 不上八维light entry · not on the 8-axis grid
国外出品foreign-made 直连不可达direct blocked 未见于备案名单not in filing list
- 能做什么What it does
- 对话问答 + 文件与图像输入 + 联网搜索 + 语音；有临时聊天（不进历史记录、不建记忆）。Chat, file and image inputs, web search and voice; Temporary Chat mode keeps conversations out of history and memory.
- 数据去哪Where your data goes
- 官方口径：个人服务（ChatGPT / Codex 等）默认可能用你的对话训练模型，可在「设置 → 数据控制」关闭「为所有用户改进模型」或在隐私门户选「请勿使用我的内容进行训练」；临时聊天不用于训练。企业侧（Team / Enterprise / API）默认不训练，API 需显式 opt-in 才共享。你主动点赞/点踩的反馈，其整段对话仍可能用于训练。Per official policy: for individual services (ChatGPT / Codex), conversations may train models by default — opt out via Settings → Data Controls (turn off "Improve model for everyone") or the privacy portal ("do not train on my content"). Temporary Chats are never used for training. Business products (Team / Enterprise / API) are not trained on by default; API customers must explicitly opt in. Feedback you submit (thumbs up/down) may train models on the whole related conversation.
- 适合谁Who it suits
- 要「什么都能问一点」、接受云端处理的人。敏感内容先关训练开关或用临时聊天；企业数据走 Team / Enterprise / API 线。For general-purpose question answering that accepts cloud processing. For sensitive content, turn training off or use Temporary Chat; business data belongs on the Team / Enterprise / API line.
多少钱：价格常变，本站不存报价，请看 官方定价页。Price: prices change often; this site stores no quotes — see the official pricing page.
出处：2 条官方链接 · 核验 2026-10-20 · verifiedSources: 2 official links · checked 2026-10-20 · verified
Claude（对话应用）Claude (chat app)
Anthropic · 轻档案 · 不上八维light entry · not on the 8-axis grid
国外出品foreign-made 直连不可达direct blocked 未见于备案名单not in filing list
- 能做什么What it does
- 对话问答 + 长文档分析 + Artifacts 代码与内容产出；与本站的 Claude Code（编码 agent）是两条产品线。Chat, long-document analysis and Artifacts for code and content — a different product line from Claude Code, the coding agent.
- 数据去哪Where your data goes
- 官方隐私政策（2026-09-10 生效）：消费者对话默认可能用于训练，可在账户设置的数据隐私控制里 opt out；即使退出，被安全审查标记的对话仍会被用。删除的对话立即从前端移除、后端 30 天内自动删除。API / Enterprise 数据不适用本政策，由客户协议管辖。Per the official privacy policy (effective 2026-09-10): consumer conversations may train models by default — opt out in account settings under data privacy controls. Even after opting out, conversations flagged for safety review are still used. Deleted conversations disappear from history immediately and are removed from the backend within 30 days. API / Enterprise data is governed by customer agreements, not this policy.
- 适合谁Who it suits
- 重长文阅读与写作质量的人。要「对话绝不进训练」就在设置里 opt out；要合同级数据条款走 API / Enterprise。For people who weight long-document reading and writing quality. To keep conversations out of training, opt out in settings; for contractual data terms use API / Enterprise.
多少钱：价格常变，本站不存报价，请看 官方定价页。Price: prices change often; this site stores no quotes — see the official pricing page.
出处：1 条官方链接 · 核验 2026-10-20 · verifiedSources: 1 official links · checked 2026-10-20 · verified
Gemini（对话应用）Gemini (chat app)
Google · 轻档案 · 不上八维light entry · not on the 8-axis grid
国外出品foreign-made 直连不可达direct blocked 未见于备案名单not in filing list
- 能做什么What it does
- 对话问答 + Google 生态联动（搜索、Workspace、Chrome）+ 多模态输入；Connected Apps 可接第三方工具（含 MCP server 工具）。Chat with deep Google integration (Search, Workspace, Chrome) and multimodal input; Connected Apps can include third-party tools such as MCP server tools.
- 数据去哪Where your data goes
- 官方 Gemini Apps 隐私声明（2026-10-13 更新）：Keep Activity 开启时，你的提示、生成内容与反馈会用于改进 Google 服务（含训练生成式模型）；关闭后未来对话不用于训练（除非你提交反馈）。临时聊天与关闭活动记录时保留 72 小时、不用于训练。官方明确：不直接拿 Gmail / Drive / Photos 原库训练，但回答中产生的摘要、摘录与推断会被用于训练。人工审核员会审阅部分数据——官方提醒勿输入机密信息。Per the official Gemini Apps Privacy Notice (updated 2026-10-13): with Keep Activity on, your prompts, generated content and feedback are used to maintain and improve Google services, including training generative models; with it off, future chats are not used for training unless you submit feedback. Temporary chats and activity-off chats are retained 72 hours and not used for training. Google states it does not train directly on your Gmail / Drive / Photos libraries, but summaries, excerpts and inferences produced while answering do train models. Human reviewers review some data — the notice warns against entering confidential information.
- 适合谁Who it suits
- 已在 Google 生态里工作的人（联动是真优势）。在意训练就关 Keep Activity；在意机密内容按官方提醒不要输入。Best for people already working inside the Google ecosystem — the integration is a real advantage. Turn Keep Activity off if training use matters; per the official warning, keep confidential content out.
多少钱：价格常变，本站不存报价，请看 官方定价页。Price: prices change often; this site stores no quotes — see the official pricing page.
出处：2 条官方链接 · 核验 2026-10-20 · verifiedSources: 2 official links · checked 2026-10-20 · verified
GrokGrok
xAI · 轻档案 · 不上八维light entry · not on the 8-axis grid
国外出品foreign-made 直连不可达direct blocked 未见于备案名单not in filing list
- 能做什么What it does
- 对话问答 + 实时网络搜索 + 公开 X 帖子检索；风格可调，图像生成与编辑。Chat with real-time web search and public X post retrieval; adjustable persona, image generation and editing.
- 数据去哪Where your data goes
- 官方 Consumer FAQ：默认可能用你的内容与 Grok 回复训练模型——在 App「Settings → Data Controls → Improve the model」或 grok.com「Settings → Data → Improve the Model」关闭；Private Chat 内容不用于训练；企业与 API 内容不用于训练。未登录使用时，官方称可匿名保留内容，部分地区（EU/UK 除外）没有退出训练的选项。删除的数据 30 天内从系统移除。在 X 内使用 Grok 适用 X 的隐私政策，是另一套条款。Per the official Consumer FAQ: content and Grok's responses may train models by default — toggle it off in the app (Settings → Data Controls → Improve the model) or on grok.com (Settings → Data → Improve the Model). Private Chat content is never used for training; business and API content is not used for training. When unauthenticated, xAI says your content may be retained anonymously, and in some regions (excluding the EU/UK) there is no training opt-out option at all. Deleted data is removed within 30 days. Grok inside X falls under X's own privacy policy — a different set of terms.
- 适合谁Who it suits
- 要接 X 实时信息的人。注意两处硬边界：未登录可能没有退出训练的选项；在 X 内用是 X 的条款，不是 xAI 的。For people who want live X information baked in. Mind two hard edges: unauthenticated use may offer no training opt-out at all, and Grok inside X is governed by X's terms, not xAI's.
多少钱：价格常变，本站不存报价，请看 官方定价页。Price: prices change often; this site stores no quotes — see the official pricing page.
出处：1 条官方链接 · 核验 2026-10-20 · verifiedSources: 1 official links · checked 2026-10-20 · verified
KimiKimi
月之暗面 Moonshot AI · 轻档案 · 不上八维light entry · not on the 8-axis grid
国产domestic 大陆可直连direct OK 已备案filed
- 能做什么What it does
- 中文对话问答 + 长文档解析与检索；另有开放平台 API（Moonshot 开放平台）。Chinese-language chat with long-document parsing and retrieval; a separate Open Platform API exists for developers.
- 数据去哪Where your data goes
- 官方数据使用与共享说明：在「经安全加密技术处理、严格去标识化且无法重新识别特定个人」的前提下，输入、指令与生成回复可能用于模型训练；退出须联系客服登记（验证身份后 5-7 个工作日），账号级统一退出、不支持按对话退出，退出后历史与新增数据都不再进训练集。数据存储于中国境内服务器。未经同意、法律要求或产品功能必需，不向第三方共享个人对话数据。Kimi Business 企业版核心承诺：企业数据不进入模型训练流程，个人与企业空间完全隔离。Per the official Data Usage and Sharing notice: subject to encryption, strict de-identification and non-re-identifiability, your inputs, instructions and generated responses may be used for model training. Opting out requires a support request (completed within 5-7 working days after identity verification); it is account-wide — per-conversation opt-out is not supported — and once registered, neither historical nor new data enters the training set. Data is stored on servers within mainland China. Personal conversation data is not shared with third parties without consent, legal requirement or product necessity. Kimi Business commits that enterprise data never enters the training pipeline, with fully isolated personal and enterprise workspaces.
- 适合谁Who it suits
- 中文长文档场景的主力。退出训练要走客服登记、不是自助开关——在意这一点要先确认；企业数据走 Kimi Business。A workhorse for Chinese long-document work. Note the opt-out is a support-desk registration, not a self-serve toggle — confirm that fits your expectations; enterprise data belongs on Kimi Business.
多少钱：价格常变，本站不存报价，请看 官方定价页。Price: prices change often; this site stores no quotes — see the official pricing page.
出处：2 条官方链接 · 核验 2026-10-20 · verifiedSources: 2 official links · checked 2026-10-20 · verified
豆包Doubao
字节跳动 · 轻档案 · 不上八维light entry · not on the 8-axis grid
国产domestic 大陆可直连direct OK 已备案filed
- 能做什么What it does
- 中文对话问答 + 文生图 + 语音对话 + 文档解析；另有火山引擎「火山方舟」提供企业级 API。Chinese-language chat, image generation, voice chat and document parsing; enterprise-grade APIs are sold separately through Volcengine Ark.
- 数据去哪Where your data goes
- 官方隐私政策 §1.10：在「经安全加密技术处理、严格去标识化且无法重新识别特定个人」的前提下，输入数据、指令与生成回复可能用于模型训练；可在「设置 → 隐私与权限 → 帮助模型改进效果」（移动端）或「设置 → 帮助模型改进效果」（PC / 网页端）关闭。API 侧是另一套口径：火山引擎《豆包模型服务协议》§4.1 —— 除非你单独同意，不使用你提交或输出的内容训练基础模型。Per the official privacy policy §1.10: subject to encryption, strict de-identification and non-re-identifiability, your inputs, instructions and generated responses may be used for model training; turn it off under Settings → Privacy & Permissions → "Help improve the model" (mobile) or Settings → "Help improve the model" (PC / web). The API line follows different terms: Volcengine's Doubao Model Service Agreement §4.1 — content you submit or receive is not used to train foundation models unless you separately agree.
- 适合谁Who it suits
- 中文日常问答与文生图。App 训练开关是自助的，API 默认不训练——两条线的条款不同，别混着理解。For everyday Chinese chat and image generation. The consumer training toggle is self-serve, while the API defaults to no training — the two lines carry different terms, so don't read one through the other.
多少钱：价格常变，本站不存报价，请看 官方定价页。Price: prices change often; this site stores no quotes — see the official pricing page.
出处：2 条官方链接 · 核验 2026-10-20 · verifiedSources: 2 official links · checked 2026-10-20 · verified
先回答一件事：你要它替你干多少事Answer one question first: how much do you want it to do for you
这一分区 27 份档案，分三个方向：23 份是装在自己机器上的（IDE 形态 11 + CLI 形态 12，其中五组是同一个产品的两种装法——Aider、Claude Code、Codex、Copilot、Cursor 各有一份 IDE 形态和一份 CLI 形态，它们不是竞品，差别在装在哪、怎么被唤起、能不能塞进脚本）；另 4 份是厂商云形态——本机什么都不装，工作在别人的机器上做。This section holds 27 entries, split three ways: 23 are installed on your own machine (11 IDE form + 12 CLI form, of which five pairs are the same product in two forms — Aider, Claude Code, Codex, Copilot and Cursor each have an IDE entry and a CLI entry; they are not competitors, and they differ in where they install, how they are invoked, and whether they drop into scripts). The other 4 are vendor cloud — nothing installs locally, the work happens on someone else's machine.
-
1我要一个自己能开的编辑器I want an editor I can launch myself看 IDE · 独立编辑器See IDE · standalone editorCursor、已改称 Devin Desktop 的 Windsurf、Zed、AWS 的 Kiro 都在这一类。Cursor, Windsurf (now renamed Devin Desktop), Zed and AWS's Kiro are all in this group.
-
2我已经在用 VS Code / Cursor 了I already use VS Code / Cursor看 IDE · 扩展形态See IDE · extension formClaude Code、Codex、Copilot、Cline、JetBrains 的 Junie、Continue 都是在既有编辑器里装。注意其中 Claude Code / Codex / Copilot 另有 CLI 形态，别当成两个产品比。Claude Code, Codex, Copilot, Cline, JetBrains' Junie and Continue all install into an existing editor. Note that Claude Code, Codex and Copilot also have CLI forms — do not compare them as if they were two products.
-
3我不想换编辑器，也不想装插件I do not want to switch editors or install plugins看 Aider（Watch 模式）See Aider (Watch mode)它不是编辑器插件而是后台常驻进程，你在任意编辑器里加 AI 注释它就响应。It is not an editor plugin but a background resident process: add an AI comment in any editor and it responds.
-
4我要它每一步都问我I want it to ask me at every step看默认逐次确认的那几个 CLISee the CLIs that confirm every action by defaultCrush 默认每次工具调用都问；Aider 有自动提交但沙箱机制未核验。Crush asks before every tool call by default; Aider has auto-commit but its sandbox mechanism is unverified.
-
5我要它在指定范围内自己跑I want it to run on its own within a set scope看有明确沙箱 / 可信目录的那几个See those with an explicit sandbox / trusted directoriesGemini CLI 的沙箱与 Trusted Folders 都有独立官方文档；Claude Code CLI 的权限粒度最细，还有「允许这一次」。Gemini CLI documents its sandbox and Trusted Folders separately; Claude Code CLI has the finest permission granularity, including "allow just this once".
-
6我要它能塞进 CI / 脚本里跑I need it to run in CI / scripts有一整批有官方证据的落点A whole batch now has official evidenceGemini CLI、Codex CLI 之外，2026-10 收录的一批也都给了非交互文档：Cursor CLI（headless）、Copilot CLI（
-p）、Continue（cn -p）、goose（goose run）、Droid（droid exec）、Kiro CLI（--no-interactive）。Beyond Gemini CLI and Codex CLI, the 2026-10 additions all document non-interactive execution: Cursor CLI (headless), Copilot CLI (-p), Continue (cn -p), goose (goose run), Droid (droid exec) and the Kiro CLI (--no-interactive). -
7我要关掉客户端后它还在干活I want it to keep working after I close the client主看厂商云形态那 4 份Start with the 4 vendor-cloud entries那一层是「整台机器都在云上」：Dot / Muse / Grok Bot / Jules，差别在审批与异步流程怎么设计——Muse 有独立的 Sentinel agent，Dot 是自动审核，Grok Bot 官方页没写，Jules 以「批准计划」为节点（API 可跳过）。本地形态里也有一批给特定任务提供云端执行（Cursor 的云端 Agent、Kiro 的 Cloud Session、Amp 的 orb、Copilot 的 coding agent），所以「关机后能不能跑」要按对象的
background维度逐个看，别按形态一刀切。That layer is 'the whole machine is in the cloud': Dot, Muse, Grok Bot and Jules — they differ in how approval and async flow are designed — Muse has a separate Sentinel agent, Dot reviews automatically, Grok Bot's official page does not say, and Jules gates on an approved plan (skippable via API). Some local forms also offer cloud execution for specific tasks (Cursor's cloud agents, Kiro's Cloud Sessions, Amp's orb, Copilot's coding agent) — so check the `background` dimension per entry instead of assuming by form. -
8我要数据完全不离开本机I need data never to leave my machine这一层也没有合适的Nothing here fits either厂商云形态的机器在别人那里（Dot 默认与本机隔离、授权后才碰）；本地形态里只有 sandbox / 可信目录那一档能限制写入范围。The vendor-cloud machines are on someone else's (Dot is isolated from your machine by default and only touches it once authorised); among local forms, only the sandbox / trusted-directory tier can restrict write scope.
⚠ 两个最容易踩的误解：「AI 原生编辑器」不等于「模型可自选」——形态看 runtime，模型接入看 model_access，两个维度是分开的；Codex CLI 的审批模式与沙箱档位在 CLI 文档里查不到，在它 SDK 的源码里有——四种审批 × 三档沙箱（read-only / workspace-write / danger-full-access），还能按路径 deny 掉读 .env，跨平台差异仍未核验。⚠ Two misconceptions people fall into most easily: "AI-native editor" does not mean "model is selectable" — form is `runtime`, model access is `model_access`, and the two dimensions are separate. Codex CLI's approval modes and sandbox tiers are not in its CLI documentation, only in its SDK source: four approval modes × three sandbox tiers (read-only / workspace-write / danger-full-access), plus per-path denies on reading `.env`. Cross-platform differences remain unverified.
按装法分By install form
同一产品常有 IDE 与 CLI 两种装法，它们不是竞品——装在哪、怎么被唤起、能不能进脚本都不同。选你实际要用的那个。A product often has both an IDE and a CLI form. They are not competitors — they differ in where they install, how they are invoked, and whether they drop into scripts. Pick the one you will actually use.
IDE 形态IDE form 11
IDE 形态IDE form装进编辑器，或本身就是独立编辑器。看得见界面，权限多在对话里给。Installed into an editor, or an editor in its own right. You see a UI, and permissions are mostly granted in conversation.
Aider Watch 模式Aider (Watch mode)
Aider-AI
不是编辑器插件，是一种使用模式。 Aider 常驻 watch 你的文件，你在任意编辑器里加 AI 注释，它就响应。Not an editor plugin but a usage mode — Aider stays resident a
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
一个进程、三种入口——终端、IDE、GitHub @claude 提及共用同一套权限与上下文机制。One process, three entry points — terminal, IDE and GitHub @cl
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
ClineCline
Cline Bot
本分区唯一三形态全覆盖的对象。CLI、Desktop App、VS Code 扩展共享同一开源核心。模型也是自选的。The only object here covering all three forms — CLI, desktop a
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Codex IDE 扩展是装进 VS Code / Cursor / Windsurf 的扩展，不是独立 IDE —— 它通Not a standalone IDE but an extension for VS Code / Cursor / W
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
ContinueContinue
Continue
开源、模型完全自配的 IDE 扩展。 但它已停更：Continue 被 Cursor 收购，上游仓库转入 read-onlyAn open-source IDE extension with fully self-configured models
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
GitHub CopilotGitHub Copilot
GitHub / Microsoft
形态最广、生态最依附 GitHub 的一个——从网站到 CLI 到主流编辑器全覆盖，且能把任务委派给 Claude / CoThe widest form coverage and the most GitHub-bound ecosystem —
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
CursorCursor
Anysphere
本分区架构最激进的三个之一： 云端 Agent 合上笔记本不中断、自托管代码不出网、代码审查拆成两个 bot 分工。One of the three most architecturally aggressive here: a cloud
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
JunieJunie
JetBrains
Junie 是 JetBrains 官方的 AI coding agent。 它的形态是插件加 CLI：插件装进 JetBrJetBrains' official "AI coding agent" — a plugin inside JetBra
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
KiroKiro
AWS
AWS 出的 agentic 工程环境：一套 harness 同时撑起 IDE / CLI / Web / Mobile 四AWS's agentic engineering environment: one unified harness acr
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
已改称 Devin Desktop（被 Cognition 收购后）——有自研模型 SWE-2、支持 ACP previewNow renamed Devin Desktop (after the Cognition acquisition) —
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
ZedZed
Zed Industries
不是「又一个 AI 编码工具」，而是三种 agent path 的编排层 —— 能在同一个编辑器里用 Zed 原生 AgenNot "yet another AI coding tool" but an orchestration layer fo
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
CLI 形态CLI form 12
CLI 形态CLI form在终端里跑。能进脚本与 CI，权限确认粒度是最细的一层。Runs in the terminal. Drops into scripts and CI, with the finest granularity of permission confirmation.
AiderAider
Aider-AI
Aider 是一个终端 pair programming 工具。 它把 Git 集成当核心设计，不是附加功能。 AI 改动始A terminal pair-programming tool where Git integration is core
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
AmpAmp
Sourcegraph
Amp 是 Sourcegraph 出的前沿编码 agent，CLI + 云端 orb + runner 三执行面。 发一句Sourcegraph's Amp: a frontier coding agent spanning CLI + clou
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Claude Code CLIClaude Code CLI
Anthropic
本分区数据质量最高、发布最频繁的条目。它自带 CHANGELOG 与 Atom feed。 它的权限设计颗粒度最细。The highest data quality and most frequent releases in this tr
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Codex CLICodex CLI
OpenAI
OpenAI 官方的终端编码 agent，Rust 实现，架构上强调 app server 与线程续接。OpenAI's official terminal coding agent, written in Rust, arch
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Copilot 在终端里的编码 agent 形态，与 GitHub 订阅同源。 它能在终端里走完一条链。 "读仓库 / 改代The Copilot agent in your terminal, carried by your GitHub sub
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
CrushCrush
Charm
crush 是 Charm 生态出品的 TUI 编码 agent，在终端里跑。 它的 MCP 三种传输方式标注最完整，并带一A TUI coding agent in the Charm ecosystem that runs in your te
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Cursor CLICursor CLI
Anysphere
Cursor 的终端形态：和编辑器共用同一账户、同一批模型、同一份 mcp.json。 命令名是 agent。 官方称它支持The terminal form of Cursor: it shares one account, one model
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
DroidDroid
Factory AI
Droid 是 Factory AI 的终端编码 agent。 它模型无关——官方托管、BYOK、本地模型三路并存。 它以滚Factory AI's terminal coding agent, model-agnostic (factory-ho
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Gemini CLIGemini CLI
Google
Google 官方的开源终端 agent，npx 免安装启动。 明确支持 MCP，另有非交互脚本模式。Google's official open-source terminal agent — npx, no install
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
GooseGoose
Agentic AI Foundation
本地优先的开源通用 agent：桌面应用 + CLI + 可嵌入 API 三种形态都能跑。 由 Block 创建，现属 AgCreated by Block and now hosted by the Agentic AI Foundation (
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
OpenCodeOpenCode
OpenCode
MIT 许可的多 provider 编程 harness。它运行 bash 命令前会询问权限，并支持 ACP 载入。An MIT-licensed multi-provider programming harness that asks p
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
WarpWarp
Warp
开源的终端 agent 环境。 官方自称： 「Agentic Development Environment」 它把现代终端An open-source Agentic Development Environment — the vendor's
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
厂商云形态Vendor cloud 4
厂商云形态Vendor cloud本机不装任何东西，在厂商的云端机器上持续工作（OpenAI Dot / Meta Muse / xAI Grok Bot）。差别不在功能多寡，而在机器归谁 —— 代价是本地文件与离线能力基本让出，收益是关机后仍在跑。Nothing installs locally; the work happens on the vendor's cloud machine (OpenAI Dot / Meta Muse / xAI Grok Bot). The difference is not how many features you get but whose machine it runs on — the cost is giving up local files and offline capability, the benefit is that it keeps running after shutdown.
它是三家同期发布里唯一一家官方页上能读到「jobs 清单」的。 它不谈机制，直接列了九个岗位：销售线索、网站搭建、收件箱、会Of the three released at the same time, it is the only one who
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
JulesJules
Google
本机什么都不装，任务跑在 Google 的云机器上，做完给你一个 PR。 它和你熟悉的 IDE / CLI 助手最大的差别不Nothing installs locally: the work runs on a Google Cloud VM a
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
Meta MuseMeta Muse
Meta
唯一一家把「审批权交给另一个 agent」的。Muse Secure VM 上跑着两个系统级隔离的东西：干活的 Muse，和The only one that hands approval authority to another agent —
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
OpenAI DotOpenAI Dot
OpenAI
OpenAI 官方的云端编码 agent：本机什么都不装，所有工作发生在 OpenAI 的云端机器上。官方给的能力边界写得很Nothing installs locally; all work happens on OpenAI's cloud m
// line 已是双节点，勿走 esc -->
部分核验Partly verified
详情 →Details →
本站目前是「核验快照」，还没有实测数据This site is currently a verification snapshot — no measured runs yet
下面每一个字段都能回到官方原文核对，但我们还没有在同一批任务上跑过这些工具。Every field below can be traced back to the official source, but we have not yet run these tools on the same batch of tasks. 实测协议已经写好（统一任务、统一验收、记录人工介入与返工次数），任务清单也已就绪，The measurement protocol is written (shared tasks, shared acceptance criteria, recording human interventions and rework counts) and the task list is ready, but it is 尚未执行。not yet run.
所以本站给的是能力边界与证据，不是「哪个更好用」的结论。So what this site gives you is capability boundaries and evidence, not a verdict on which is better to use. 真正的选型请用你自己的输入、预算与验收标准跑一遍。For an actual choice, run them yourself against your own inputs, budget and acceptance criteria.
8 个维度，每格都能回溯8 dimensions, every cell traceable
所有对象在同一坐标系下被描述。All objects are described on the same coordinate system. 点开任一对象，每一格下面都有官方源链接与核验日期——不认同可以自己回去查。Open any object and every cell carries an official source link and a checked date — disagree and verify it yourself.
01
模型与开放条件Model & openness
用谁的模型？能不能换第三方？多模型路由怎么配？Whose model does it use? Can you switch to a third party? How is multi-model routing configured?
02
运行位置Where it runs
本地 / 云端 / 混合？索引在本地还是云端？远程控制不等于云端执行。Local, cloud or hybrid? Is the index local or remote? Remote control is not the same as cloud execution.
03
本地文件Local files
能访问多大范围？未打开的大文件怎么读？符号级能力如何？How wide an area can it reach? How does it read large files you have not opened? How capable is it at the symbol level?
04
关机后的任务Tasks after shutdown
电脑关了还能继续跑吗？这是本地形态最常见的误解来源。Does it keep running once the computer is off? This is the most common misconception about local forms.
05
工具与扩展Tools & extensions
工具链有多宽：MCP · LSP · 插件 · Hooks · 自定义指令。How wide is the toolchain: MCP · LSP · plugins · hooks · custom instructions?
06
上下文与记忆Context & memory
索引算法、上下文窗口、会话恢复与分叉、跨会话记忆边界。Indexing algorithm, context window, session resume and fork, and the boundary of cross-session memory.
07
权限与限制Permissions & limits
沙箱与审批边界、凭据管理、数据是否出境、额度如何计算。Sandbox and approval boundaries, credential handling, whether data leaves your machine, and how quota is charged.
08
适合什么任务Best for
最擅长什么任务？什么情况下别用它？What tasks is it best at? When should you not use it?
全部数据一览All data at a glance
同一张表横向对比。点对象名进详情页看逐格证据。The same table side by side. Click an object name to see per-cell evidence.
| 对象Object | 形态Form | 厂商Vendor | 适合Best for | 核验日Checked | 状态Status |
|---|---|---|---|---|---|
| AAiderAider | CLI 形态CLI form | Aider-AI | 适合三类场景。 留在 git 里的改动：希望 AI 改动始终留在 git 版本控制里。 这样的改动可 diff、可撤销。 多模型接入：需要连Scenarios where AI changes must stay in git version control, diffable | 2026-10-08 | partial |
| AAider Watch 模式Aider (Watch mode) | IDE 形态IDE form | Aider-AI | 适合两类人： 习惯在任意编辑器里工作，不愿切换到专门的 Agent 界面的人 需要「加注释 → 自动响应」这种轻量交互的人People who work in whatever editor they like and refuse to move to a d | 2026-10-08 | partial |
| AAmpAmp | CLI 形态CLI form | Sourcegraph | 适合四类用法： 想让 agent 在自己的云端机器（orb）里长跑、合上笔记本仍继续的人 需要多模型前沿路由 + Oracle 第二意见的团Four kinds of use fit: People who want an agent that keeps working in | 2026-10-08 | partial |
| CCClaude Code CLIClaude Code CLI | CLI 形态CLI form | Anthropic | 重视权限可控性、需要 MCP 生态、希望用 Anthropic 官方模型的用户。 不适合需要开源许可或自托管的场景（许可明确 proprieUsers who value permission control, want the MCP ecosystem, and want A | 2026-10-08 | partial |
| CCClaude Code IDE 扩展Claude Code (IDE extension) | IDE 形态IDE form | Anthropic | 已在用 IDE 且希望不切换工具就能用 Claude 的人。 需要在终端、IDE、GitHub 三处用同一工具的人。 重视权限可控性与 MCPeople already in an IDE who want Claude without switching tools. Peop | 2026-10-08 | partial |
| CClineCline | IDE 形态IDE form | Cline Bot | 适合四类人： 需要同一工具覆盖编辑器、终端、桌面三种工作形态 需要模型自选（不想绑定单一厂商） 需要 CI/CD 集成（CLI headleUsers who need one tool across editor, terminal and desktop. People wh | 2026-10-08 | partial |
| CCodex CLICodex CLI | CLI 形态CLI form | OpenAI | 适合在终端里做工程任务，要接 CI 或要自建客户端的场景。 它有个别人少有的证据：非交互模式有官方文档支持。 docs/exec.md 的标Engineering work in the terminal, including CI integration or building | 2026-10-08 | partial |
| CCodex IDE 扩展Codex (IDE extension) | IDE 形态IDE form | OpenAI | 适合两类人： 已在用 VS Code / Cursor / Windsurf 等编辑器，希望在同一环境里用 Codex 而不切换工具的人 需A fit for two groups: People already in VS Code / Cursor / Windsurf wh | 2026-10-08 | partial |
| CoContinueContinue | IDE 形态IDE form | Continue | 适合三类人： 需要模型自由（多 provider，含 Ollama 本地模型）而不想被单一厂商锁定的人。 已在 VS Code / JetBA fit for three groups: People who need model freedom (many providers, | 2026-10-08 | partial |
| CPGitHub Copilot CLIGitHub Copilot CLI | CLI 形态CLI form | GitHub | 适合三类人： 已在 GitHub 生态、想在终端把「读仓库 / 改代码 / 跑测试 / 开 PR / 管 Issue」走完的人。官方用例直接People already in the GitHub ecosystem who want to walk the whole chai | 2026-10-08 | partial |
| GHGitHub CopilotGitHub Copilot | IDE 形态IDE form | GitHub / Microsoft | 已在 GitHub 生态内（用 PR / Issue / Actions 工作）的人； 需要最广编辑器覆盖的人（VS Code / JetBPeople already inside the GitHub ecosystem (working via PR / Issue / A | 2026-10-08 | partial |
| CRCrushCrush | CLI 形态CLI form | Charm | 适合这几类人： 喜欢 Charm 生态的终端体验（TUI 审美与交互） 需要在多个 provider 间切换，或想用官方订阅方案 HyperSeveral kinds of people fit: Users who like the Charm ecosystem's term | 2026-09-29 | partial |
| CCursor CLICursor CLI | CLI 形态CLI form | Anysphere | 适合三类人： 在终端里做工程任务、写脚本、或进 CI/CD 的人（官方 GitHub Actions 文档与 cookbook） 已用 CuPeople doing engineering work in the terminal, writing scripts, or run | 2026-10-08 | partial |
| CCursorCursor | IDE 形态IDE form | Anysphere | 适合三类人： 需要云端 Agent、希望合上笔记本任务继续跑的人 需要长周期（数月级）上下文维持与大规模子智能体编排的人 代码不能离开自有基People who want a cloud agent that keeps running when they close the l | 2026-10-03 | partial |
| DDroidDroid | CLI 形态CLI form | Factory AI | 适合在终端里跑端到端软件工程、且需要企业级治理的团队。 三类团队最对得上： 已有仓库，想跑完规划 → 实现 → 测试的完整流程（/missiEnd-to-end software engineering in the terminal for teams that need en | 2026-10-08 | partial |
| GGemini CLIGemini CLI | CLI 形态CLI form | 适合三类场景： 想免安装快速试用（npx，不留安装痕迹） 需要 MCP 扩展，尤其是把 Gemini 的多模态生成能力（Imagen / VThree scenarios fit: You want a quick trial without installing anythin | 2026-09-29 | partial | |
| GGooseGoose | CLI 形态CLI form | Agentic AI Foundation | 四类人适合用： 想要一个本地优先、可换任意 provider（含本地模型）的通用 agent，而不止写代码的人 需要桌面应用 + CLI 双A fit for four kinds of people. People who want a local-first, general | 2026-10-08 | partial |
| xAIxAI Grok BotxAI Grok Bot | 厂商云形态Vendor cloud | xAI | 适合按岗位选人的人。 官方给的是一份岗位化的 jobs 清单（见 tools 维），可以当清单用：销售线索、 网站搭建、邮箱清理、客服、游戏A fit if you pick tools by job to be done. The official list of nine j | 2026-10-03 | partial |
| JLJulesJules | 厂商云形态Vendor cloud | 官方给自己划的活是具体的：把任务描述交进 GitHub 仓库后，它做这些事： 官方原文："fix bugs, update dependenSelf-contained, verifiable work: fixing a named bug, bumping dependenc | 2026-10-08 | partial | |
| JJunieJunie | IDE 形态IDE form | JetBrains | 适合三类人，不适合另外三类。 适合： 已经在用 JetBrains IDE（IntelliJ IDEA / PyCharm / WebStoPeople already using JetBrains IDEs (IntelliJ IDEA / PyCharm / WebStor | 2026-10-08 | partial |
| KKiroKiro | IDE 形态IDE form | AWS | 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 sessioTeams doing formal engineering delivery in the AWS ecosystem who want | 2026-10-08 | partial |
| MUMeta MuseMeta Muse | 厂商云形态Vendor cloud | Meta | 官方定位是 官方原文："for billions of people worldwide, so there's no learning cFits (official positioning): everyday tasks plus longer goals — bookin | 2026-10-03 | partial |
| OAOpenAI DotOpenAI Dot | 厂商云形态Vendor cloud | OpenAI | 官方给的是岗位化场景，值得直接抄成选型依据： 把反馈转成经过测试的修复（交完整 PR 附演示视频） 产品范围变化时调整发布方案 拿到新证据后Fits (scenarios the vendor itself gives, usable directly as selection | 2026-10-03 | partial |
| OCOpenCodeOpenCode | CLI 形态CLI form | OpenCode | 需要多 provider 可切换、且重视命令执行前确认的用户。 ACP 生态参与者。Users who need switchable multiple providers and who care about confir | 2026-10-08 | partial |
| WWarpWarp | CLI 形态CLI form | Warp | 适合：在终端里同时做命令执行与编码 agent 工作、并希望团队协作与云自动化的用户。 特别适合： 想把终端工作流标准化为 WorkflowDevelopers who run both shell commands and coding agent work in the te | 2026-10-08 | partial |
| WWindsurf / Devin DesktopWindsurf / Devin Desktop | IDE 形态IDE form | Cognition | 需要多模型供应商可选（不想被单一模型绑定）的人； 需要 ACP 以接入自定义 agent 的人（注意是 preview）； 需要远程主机（SPeople who want a choice of model providers rather than being locked t | 2026-10-08 | partial |
| ZZedZed | IDE 形态IDE form | Zed Industries | 适合重视编辑器性能（Rust）、解析准确度（Tree-sitter）与多人协作的场景。 也适合想在一个编辑器里混用多种 CLI Agent Scenarios where editor performance (Rust), syntactic parsing accuracy | 2026-10-08 | partial |
为什么这里不给总分排名Why there is no overall ranking
这不是省略，是方法上的明确取舍。This is not an omission but a deliberate methodological choice.
- 缺统一实测，就不给分。跨工具的跑分口径不同——题目、预算、执行环境都不一样，A 的 90 分和 B 的 88 分不可比。No shared measurement, no score. Benchmarks across tools use different rules — different questions, budgets and environments. A 90 and a B 88 are not comparable.
- 「未知」是合法答案。"Unknown" is a legitimate answer.查不到就写「未知」并说明为什么，不用推测填充。缺信息本身也是信息。
- 来源链接、类型与核验日都在详情页里，可以逐条回去复核。The source link, its kind and the checked date are all on the detail page, so you can walk any of them back and re-check.
- lifecycle 单独标注。核验日新不等于数据有效——上游归档后数据会失效，所以单独标
active/maintenance/archived。lifecycle is marked separately. A recent check date does not mean the data is still valid — it goes stale when upstream archives the repo, soactive/maintenance/archivedare tracked separately. - 不同层不混排。成品 agent、自己搭的底座、给 agent 装的工具是三种角色，坐标系不同；硬合成一张 33 行的表只会制造假的可比性。Layers are not mixed. Finished agents, bases you assemble yourself, and tools you install for an agent are three different roles with different coordinate systems; forcing them into one 33-row table would manufacture false comparability.
可信度标记怎么读How to read the confidence marks
详情页与上表都带Both the detail pages and the table above carry confidence，它反映, which reflects 本站数据的完整度how complete this site's data is，不是对产品的评价。, not a judgement of the product.
| verified | 8 个维度均有官方源支撑，且核验日在 90 天内all 8 dimensions have official sources and were checked within 90 days |
|---|---|
| partial | 部分维度标为未知，或官方文档不可访问，或核验日超过 90 天some dimensions are marked unknown, or the official docs were unreachable, or checked over 90 days ago |
| stale | 官方已发布重大变化，本站尚未核验the vendor has shipped a significant change that this site has not re-checked |
本站当前的完整度Current completeness on this site
数据层用 node scripts/audit-gaps.mjs 可随时复核这个数字。
以下是最近一次核验的快照（核验日 2026-09-29）：
| 形态Form | 已补齐Filled | 说明Meaning |
|---|---|---|
| agents | 211/216 | 多为闭源产品，索引策略等细节官方不公开；CLI 类官方文档相对完整Mostly closed-source, so details like indexing strategy are not disclosed; CLI entries have better official docs |
全站合计 415/446 维度已补齐Site-wide 415/446 dimensions filled（93%）
剩余未补齐项分三类：官方未公开（索引算法、沙箱实现细节本就不对外说明）、需实测才能确定（大仓库表现、CI 无 TTY 行为）、客观渠道不可达。The remaining gaps fall into three kinds: not disclosed by the vendor (indexing algorithms, sandbox internals are simply not documented), requires measurement (behaviour on very large repositories, CI without a TTY), and unreachable through objective channels.
我们选择留白而不是填「已支持」——错误的成本最终由使用者承担。We chose to leave a blank rather than fill in "supported" — the cost of an error is ultimately borne by whoever uses it.
数据来源与复核方式Sources and how to re-check them
全部数据来自开源仓库All data comes from the open-source repository speculcom/ai-agent-guide（CC BY 4.0）。
每个条目都是纯 Markdown + frontmatter，含 8 个维度、证据链Every entry is plain Markdown plus frontmatter, carrying 8 dimensions and an evidence chain接、核验日与可信度标记。仓库内含：
METHODOLOGY.md—— 方法论总纲与四条核心规则axes/—— 每个维度的定义与判定标准（含正反例）SCHEMA.md—— 数据格式规范与构建校验规则tracks/*/tasks/_protocol.md—— 实测协议(尚未执行not yet run，故本站暂无实测结论, so this site has no measured conclusions)
发现错误或有新证据，欢迎提 Issue 或 PR——每条修正都会注明依据与影响范围。Found an error or have new evidence? Open an issue or a PR — each correction notes its basis and blast radius.