← Coding

Replit AgentReplit Agent

Replit

闭源Closed source 云IDECloud IDE 自然语言建站Natural-language site building checkpointcheckpoint 后台任务Background tasks partial lifecycle: active 核验Checked 2026-10-30

Replit Agent 是「从自然语言到能上线的应用」的云 IDE agent：建项目、装基础设施、测试、部署一条龙， Free/Power/Max 模式选档不选裸模型，checkpoint 兜底回滚。

Replit Agent is the cloud-IDE agent from plain language to a deployable app: it builds the project, sets up infrastructure, tests, and deploys in one chain; Free/Power/Max modes pick tiers rather than bare models, with checkpoints as the rollback safety net.

适合与不适合When to use it, when not to

适合：想法直达可部署应用、非工程师原型、多产物同后端。不适合：以你现有代码库为主战场、要模型完全自选。

Fits: idea straight to a deployable app (database/deploy included), non-engineer prototypes, multiple artifacts sharing one backend. Not a fit: your existing codebase as the main battlefield; fully self-selecting models.

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 2026-10-21

出品属地Origin: 国外出品foreign-made　Replit（美国）。Replit (US).
国内使用Direct access: 大陆可直连direct OK　replit.com 与 docs.replit.com 大陆直连实测 200（2026-10-21）。replit.com and docs.replit.com measured 200 direct from mainland (2026-10-21).
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（2026-10-21 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 2026-10-21); 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: 按「模式」选而非按裸模型选：官方文档列 Free/Auto/Power/Max 四态——Free 是套餐额度内的免费档， Auto 自动路由，Power/Max 是付费档（Max 使用官方定位的前沿模型）；具体每档挂哪些模型官方按套餐页口径给，未逐条清点。企业工作区只有一个 Auto 模式，管理员可用 Enterprise model controls 管模型访问。
Choose by "mode" rather than by bare model: the official docs list four modes — Free/Auto/Power/Max. Free is the free tier within plan allowances, Auto routes automatically, Power/Max are paid tiers (Max uses the strongest frontier models). Which models sit behind each tier follows the official plan-page framing; not enumerated one by one. Enterprise workspaces have only the Auto mode; admins can manage model access via Enterprise model controls.
运行位置Where it runs: 云端 IDE 内建 agent：项目、依赖、数据库、部署全在 Replit 云上；官方文档明确 Agent 会「写代码、搭基础设施、测结果」并自动建 checkpoint 可回滚。桌面/手机是入口，执行在云上。
Built-in agent in a cloud IDE: projects, dependencies, databases and deployments all live on Replit cloud. The official docs state the Agent "writes code, sets up infrastructure, tests results" and automatically creates checkpoints you can roll back to. Desktop/mobile are entry points; execution happens in the cloud.
本地文件Local files: 代码与数据默认在 Replit 云项目里，不在你本机；产物可导出（官方有导出/迁移线，细节未逐条核验）。连接外部服务（BigQuery/Slack/Notion/Linear 等）经官方 connector 从聊天直接取数。
Code and data default to living in Replit cloud projects, not on your machine; artifacts can be exported (official export/migration lines exist; details not verified one by one). External services (BigQuery/Slack/Notion/Linear etc.) are connected via official connectors, pulling data directly from chat.
关机后的任务Tasks after shutdown: 有后台任务面：官方文档给 Core 1 个、Pro 10 个「Active background tasks」的配额行—— 后台跑任务是产品语义的一部分；Plan mode 先出任务清单供审批再动工。
There is a background-task surface: the official docs give quota lines of Core 1 / Pro 10 "Active background tasks" — background tasks are part of the product semantics. Plan mode produces a task list for approval before work starts.
工具与扩展Tools & extensions: agent 自带建项目/装依赖/测试/部署全链；connector 面（BigQuery/Slack/Notion/Linear 等）是官方列出的连接目录，未逐条清点。MCP 支持状态未核验到官方声明。
The agent ships with the full chain — create projects, install dependencies, test, deploy. The connector surface (BigQuery/Slack/Notion/Linear etc.) is an officially listed connection directory; not enumerated one by one. MCP support status: not verified to an official statement.
上下文与记忆Context & memory: 项目内多产物共享同一后端与数据（web/mobile/slides/视频同项目）；checkpoint 提供状态回滚面。跨项目记忆未见官方声明。
Multiple artifacts within a project share the same backend and data (web/mobile/slides/video in one project); checkpoints provide a state-rollback surface. Cross-project memory: no official statement seen.
权限与限制Permissions & limits: 账号+套餐体系；付费动作需确认（官方 FAQ：paid actions require confirmation before they start， Free 不够时给 Allow once 单次放行）；企业版有模型访问管控。部署即 Replit 托管 URL。
Account + plan system. Paid actions require confirmation (official FAQ: "paid actions require confirmation before they start"; when the Free tier runs out it offers "Allow once" for one-time approval). Enterprise has model-access controls. Deployment yields a Replit-hosted URL.
适合什么任务Best for: 适合：从自然语言直接到「能上线的应用」（含数据库/部署）、非工程师做原型、要 checkpoint 回滚兜底。不适合：要接你自己已有的代码库为主战场（它主轴是从想法到新项目）、要模型层完全自选。
Good fit: going from natural language straight to a deployable app (database and deployment included), non-engineers building prototypes, wanting checkpoint rollback as a safety net.

Not a fit: making your existing codebase the main battlefield (its main axis is idea → new project), wanting full self-selection at the model layer.

头号误解Top misconception

把 Replit Agent 当「装进现有仓库的编码 agent」——官方定位是从想法到新项目（含基础设施），现有代码库协作不是它文档的主叙事Treating Replit Agent as 'a coding agent installed into your existing repo' — the official positioning is idea-to-new-project (infrastructure included); collaborating on existing codebases is not the docs' main narrative
Free/Power/Max 是「模式」不是模型名——每档背后挂什么模型随官方调整，别把档位当固定模型清单Free/Power/Max are modes, not model names — which models sit behind each tier shifts with official changes; don't treat tiers as a fixed model list
产物与数据默认在 Replit 云上——导出与迁移线存在但边界要自己核（未逐条核验）Artifacts and data live in Replit's cloud by default — export/migration paths exist but the boundaries are yours to verify (not verified line by line)

未知项清单Known unknowns

各模式背后具体模型清单（官方按套餐页口径，未逐条清点）
MCP 支持状态（未核验到官方声明）
产物导出与数据迁出的细节边界（未逐条核验）
跨项目记忆机制（未见官方声明）

The concrete model list behind each mode (official framing follows the plan pages; not enumerated one by one)
MCP support status (not verified to an official statement)
Detailed boundaries of artifact export and data migration (not verified one by one)
Cross-project memory mechanism (no official statement seen)

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 2026-10-30，内容更新日content updated on 2026-10-21。

类型Type	名称Name	链接Link
official	Replit Docs · Replit Agent 总览（模式/Plan/后台任务/FAQ）Replit Docs · Replit Agent overview (modes / Plan / background tasks / FAQ)	https://docs.replit.com/features/agent/overview
changelog	Replit Docs · Updates（周更 changelog，最新 2026-10-16）Replit Docs · Updates (weekly changelog, latest 2026-10-16)	https://docs.replit.com/updates
official	Replit · 官网Replit · official site	https://replit.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）：

Free 档五小时重置额度的实际可用量——官方只给「allowances reset every five hours」口径
Allow once 单次放行的确认交互细节（何时弹、弹几次）
产物导出/迁出的完整路径（官方有迁移线，边界未逐条核验）

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):

The actual usable amount of the Free tier's five-hour-reset allowance — the official page only gives the framing "allowances reset every five hours"
The confirmation interaction details of "Allow once" (when it prompts, how many times)
The complete path for exporting/migrating artifacts out (official migration lines exist; boundaries not verified one by one)