← Coding

Augment CodeAugment Code

Augment Code

海外Outside China 企业Enterprise 云端AgentCloud agent CLICLI IDE扩展IDE extension MCPMCP 自研模型路由In-house model routing 被收购过渡Acquisition transition partial lifecycle: active 核验Checked 2026-10-22

Augment Code 是企业级「软件工厂」平台：Cosmos 把规划-实现-评审-反馈连成可重复的 agent 流水线， Environments 可托管云/自有基础设施/本机三选一，Code Context Engine 招牌可经 MCP 插进别的 agent，提供多厂商模型清单（含自研 Prism 路由）——2026-10-08 官方公告 Cosmos/Auggie CLI/Context Engine 资产加入 Harness。

Augment Code is an enterprise 'software factory' platform: Cosmos chains plan-implement-review-feedback into repeatable agent loops, Environments run on Augment-hosted cloud, your own infrastructure (self-hosted daemons) or your laptop, the Code Context Engine plugs into other agents via MCP, and the model menu is among the broadest (including the proprietary Prism router) — per the 2026-10-08 official notice, Cosmos, Auggie CLI and the Context Engine assets are joining Harness.

适合与不适合When to use it, when not to

适合：企业级跨 SDLC 的 agent 流水线、self-hosted 把 agent 跑在自己基础设施、要多前沿模型清单与 token 透明计费。不适合：找个人免费档入门、要产品归属完全稳定（正处 Harness 并入过渡期）。

Fits: enterprise cross-SDLC agent pipelines, self-hosted environments running agents on your own infrastructure, a broad frontier-model menu with transparent token-based billing. Not a fit: looking for a personal free tier on-ramp (the mainline is enterprise delivery with token billing); needing fully stable product ownership (mid-transition into Harness).

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 2026-10-22

出品属地Origin: 国外出品foreign-made　Augment Code（美国，企业级 AI 编码公司）；2026-10 起 Cosmos/Auggie CLI/Context Engine资产并入 Harness（同为美国企业软件公司）。Augment Code (US enterprise AI-coding company); since 2026-10 its Cosmos, Auggie CLI and Context Engine assets are joining Harness (also a US enterprise software company).
国内使用Direct access: 大陆可直连direct OK　augmentcode.com / docs.augmentcode.com 大陆直连实测 200（2026-10-22）。augmentcode.com / docs.augmentcode.com measured 200 direct from mainland (2026-10-22).
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（2026-10-22 检索）；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 (searched 2026-10-22); 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: 官方 Available Models 页列全清单：Anthropic Claude 系（Fable 5.1/5、Opus 5/4.8/4.7/4.6、Sonnet 5.5/5、Haiku 4.5）、 Google Gemini（3.8/3.7 Flash、3.1 Pro）、OpenAI GPT（6.1 Sol、6 Astra、5.6 Sol/Terra/Luna、5.5）、 xAI Grok 4.6/4.5、智谱 GLM 5.3 Flash/5.2 与 Moonshot Kimi K3/K2.7 Code（托管在 Fireworks/Baseten）、自研 Prism（组合 Claude 与 GPT 的模型集成，Prism 选项按策展模型族路由）。计费走 token 制（Token-Based Pricing 页：LLM token + Cosmos 算力 + 服务费三段）。
The official Available Models page lists the full roster: the Anthropic Claude family (Fable 5.1/5, Opus 5/4.8/4.7/4.6, Sonnet 5.5/5, Haiku 4.5), Google Gemini (3.8/3.7 Flash, 3.1 Pro), OpenAI GPT (6.1 Sol, 6 Astra, 5.6 Sol/Terra/Luna, 5.5), xAI Grok 4.6/4.5, Zhipu GLM 5.3 Flash/5.2 and Moonshot Kimi K3/K2.7 Code (hosted on Fireworks/Baseten), plus in-house Prism (a model ensemble combining Claude and GPT; the Prism option routes across curated model families). Billing is token-based (Token-Based Pricing page: three segments — LLM tokens + Cosmos compute + service fee).
运行位置Where it runs: 主线是 Cosmos 平台（官方 Introduction：Environments/Experts/Sessions/Capabilities/Triggers 五积木）：把规划-实现-评审-反馈连成可重复的 Software Factory 工作流（Requirements→Ticket、Ticket→PR、 PR→Merge、Signal→Ticket 四条官方 loop）。Environments 三选一：Augment 托管云沙箱、你自己的 VM/物理机（self-hosted daemon）、甚至本机笔记本。另有 Auggie CLI（终端 agent：仓库实现/代码探索/跑测试/CI 自动化）与 VS Code / JetBrains 的 IDE 扩展面。
The main line is the Cosmos platform (official Introduction: five building blocks — Environments/Experts/Sessions/Capabilities/Triggers): it chains plan-implement-review-feedback into a repeatable Software Factory workflow (four official loops: Requirements→Ticket, Ticket→PR, PR→Merge, Signal→Ticket). Environments come in three flavors: Augment-hosted cloud sandbox, your own VM/physical machine (self-hosted daemon), or even your local laptop. There is also the Auggie CLI (terminal agent: repo implementation / code exploration / running tests / CI automation) and IDE extension surfaces for VS Code / JetBrains.
本地文件Local files: IDE 扩展与 Auggie CLI 处理本机代码库；Cosmos 的 Environments 决定代码在哪——托管云沙箱、 self-hosted daemon（跑在你自己基础设施上，daemon-config.json 配工作区与隔离）或本机。 self-hosted 面是官方一等公民（官方页明示 own virtual or physical infrastructure）。
IDE extensions and the Auggie CLI work on local codebases; Cosmos's Environments decide where the code lives — hosted cloud sandbox, self-hosted daemon (running on your own infrastructure, configured via daemon-config.json for workspace and isolation) or local. The self-hosted surface is an official first-class citizen (the official page explicitly says own virtual or physical infrastructure).
关机后的任务Tasks after shutdown: Cosmos 主打长时多步交付：durable environments、scheduled automations（定时自动化）、跨仓库跨工具协调（官方 llms.txt 使用指引原文）；Triggers 把事件接进 loop。官方叙事「always-on software delivery system」——从 spec 到验证常跑。
Cosmos leads with long-horizon multi-step delivery: durable environments, scheduled automations, cross-repo cross-tool coordination (wording from the official llms.txt usage guide); Triggers wire events into the loop. The official narrative is an "always-on software delivery system" — running continuously from spec to verification.
工具与扩展Tools & extensions: MCP 面官方成页多处：Cosmos Experts 接 MCP servers（MCP Registry）、CLI/IDE 侧 MCP 配置、 Context Engine MCP（把上下文引擎以 MCP 插进任何 agent）、Code Review 的 MCP context 接入；官方还提供 Docs MCP server（Streamable HTTP）给外部 agent 查 Augment 文档。
The MCP surface has multiple official pages: Cosmos Experts connect MCP servers (MCP Registry), CLI/IDE-side MCP configuration, Context Engine MCP (plugging the context engine into any agent via MCP), MCP context hookup for Code Review; the official side also ships a Docs MCP server (Streamable HTTP) so external agents can query Augment docs.
上下文与记忆Context & memory: 招牌是 Code Context Engine：官方口径 world-class context engine，跨代码库的结构化理解；以 MCP 形式输出给别的 agent 用（Context Engine MCP 页）。 Wisdom 类上下文（PR 历史、业务需求）在 Software Factory 各 loop 页有展开。
The signature is the Code Context Engine: the official stance is a world-class context engine with structured cross-codebase understanding; it is exported as MCP for other agents to use (Context Engine MCP page). Wisdom-class context (PR history, business requirements) is expanded on the Software Factory loop pages.
权限与限制Permissions & limits: 企业定位（enterprise-grade 口径贯穿官方文档）：Trust Center、Status Page 官方入口在导航； Analytics API 用 scoped OAuth token 或 service-account bearer token，写权限按需申请（官方 llms.txt 口径）。个人侧权限模型未逐条核验。
Enterprise positioning (the enterprise-grade stance runs through the official docs): Trust Center and Status Page entries sit in the navigation; the Analytics API uses scoped OAuth tokens or service-account bearer tokens, with write permissions requested on demand (per the official llms.txt). The individual-side permission model was not verified item by item.
适合什么任务Best for: 适合：企业级「软件工厂」叙事——要跨 SDLC 的 agent 流水线（ticket→PR→merge→monitor）、 self-hosted 环境把 agent 跑在自己基础设施、多前沿模型清单（含自研 Prism 路由）。不适合：找个人免费档入门（主线是企业交付与 token 计费）；要产品归属稳定的读者——2026-10 起 Cosmos/Auggie CLI/Context Engine 资产已并入 Harness。
Good fit: the enterprise "software factory" narrative — agent pipelines across the SDLC (ticket→PR→merge→monitor), self-hosted environments running agents on your own infrastructure, one of the fullest frontier-model rosters (with in-house Prism routing). Not a fit: looking for a personal free tier to get started (the main line is enterprise delivery with token billing); readers who need product ownership to be stable — since 2026-10 the Cosmos/Auggie CLI/Context Engine assets have been joining Harness.

头号误解Top misconception

2026-10-08 官方博客「Augment Code's next chapter」：Cosmos、Auggie CLI、Code Context Engine 及相关技术加入 Harness，团队并入——产品连续性走 Harness 体系，检索时 Augment/Harness 两个名字都要认The 2026-10-08 official blog 'Augment Code's next chapter': Cosmos, Auggie CLI, the Code Context Engine and related technology are joining Harness, with the team moving along — product continuity runs through Harness; recognize both Augment and Harness names when searching
官网与 docs 仍活跃在售 Cosmos（2026-10-22 实测 200）——「被收购」不等于「下线」，但后续归属与品牌走向官方未给完整说明The website and docs are still live and selling Cosmos (measured 200 on 2026-10-22) — 'acquired assets' does not mean 'shut down', but the official notice gives no complete roadmap for brand and ownership going forward
计费是 token 制三段（LLM token + Cosmos 算力 + 服务费）——与按 credit 包月的同类工具口径不同，对比成本前先核 Token-Based Pricing 页Billing is three-part token-based (LLM tokens + Cosmos compute + service fee) — different from monthly-credit tools; check the Token-Based Pricing page before comparing costs

未知项清单Known unknowns

Harness 并入后的品牌与产品线走向（官方公告未给完整说明）
个人/小团队计划是否存在（导航以企业销售为主，未核验到个人档）
托管云沙箱的代码存储区域与保留策略（未逐条核验）
个人侧权限模型细节（未逐条核验）

Brand and product-line direction after the Harness integration (the official announcement gives no complete statement)
Whether personal/small-team plans exist (navigation is enterprise-sales-led; no personal tier verified)
Code storage region and retention policy of the hosted cloud sandbox (not verified item by item)
Details of the individual-side permission model (not verified item by item)

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 2026-10-22，内容更新日content updated on 2026-10-22。

类型Type	名称Name	链接Link
official	Augment Code 文档 · Introduction（Cosmos 五积木/IDE 与 CLI 入口）Augment Code Docs · Introduction (Cosmos five building blocks / IDE and CLI entry points)	https://docs.augmentcode.com/introduction
official	Augment Code 文档 · Available Models（全模型清单与 Prism 路由）Augment Code Docs · Available Models (full model roster and Prism routing)	https://docs.augmentcode.com/models/available-models
official	Augment Code 文档 · Token-Based Pricing（三段计费口径）Augment Code Docs · Token-Based Pricing (three-segment billing stance)	https://docs.augmentcode.com/models/token-based-pricing
official	Augment Code 文档 · Self-Hosted Environments（自有基础设施跑 agent）Augment Code Docs · Self-Hosted Environments (agents on your own infrastructure)	https://docs.augmentcode.com/cosmos/environments/daemons
official	Augment Code 文档 · Software Factory Overview（四条官方 loop）Augment Code Docs · Software Factory Overview (the four official loops)	https://docs.augmentcode.com/cosmos/software-factory/overview
official	Augment Code 博客 · Augment Code's next chapter（2026-10-08 资产加入 Harness 公告）Augment Code blog · Augment Code's next chapter (2026-10-08 announcement of assets joining Harness)	https://www.augmentcode.com/blog/augment-codes-next-chapter
changelog	Augment Code · Changelog（changelog）Augment Code · Changelog (changelog)	https://www.augmentcode.com/changelog
official	Augment Code 官网（Cosmos/软件工厂叙事）Augment Code official site (Cosmos / software-factory narrative)	https://www.augmentcode.com

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（三条，都指向官方可读页没给的东西）：

Harness 并入后购买/续约路径的实际变化（公告只说资产与团队并入，未给客户迁移细则）
self-hosted daemon 在真实 CI 里的隔离与审计行为（官方有 daemon-config 页，未做实测）
Prism 路由的实际模型选择透明度（能否看到每次请求落到哪个模型）

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

Suggested measurement (three, each pointing at something the official readable pages do not give):

Actual changes to the purchase/renewal path after the Harness integration (the announcement only covers assets and team joining; no customer migration details)
Isolation and audit behavior of the self-hosted daemon in real CI (there is an official daemon-config page; not measured)
Actual transparency of Prism routing's model choice (can you see which model each request lands on)