Replit Cloud Agents 对 Pro/Enterprise 开放:可对现有公司代码库提已测 PR
二手报道称 Replit 已向全部 Pro 与 Enterprise 客户开放 Cloud Agents 指向现有公司代码库:用自然语言描述改动,智能体在副本中预览并测试后,向 GitHub/GitLab 开带截图与视频的 PR,工程师审批;亦可在 Slack 提及触发。
很多「AI 写应用」工具止步于从零生成玩具项目,改不动公司主仓。据开发者转述,Replit 已向全部 Pro 与 Enterprise 客户开放:Cloud Agents 可对准现有公司代码库。团队里任何人用自然语言描述产品改动(例如「修好账单页间距」),智能体定位相关代码、给出运行中应用的实时预览,在应用副本里点击自测并修复失败路径,然后向 GitHub 或 GitLab 开出带截图与视频的拉取请求(PR),由工程师审批合并。管理员一次性配好环境后,也可在 Slack 里提及 Replit 发起请求。
报道强调这是从「旁路原型」迈向「动核心产品」的一步:懂问题的设计、支持或运营同学可以交出可运行的变更草案,但不绕过工程师评审。对软件研发智能化,价值在于把自然语言变更接到既有 PR 门禁,而不是另开影子仓库。原帖未披露仓库规模上限、测试沙箱隔离级别、计费增量与合规认证范围;能力以 Replit 控制台与文档为准,本稿为二手整理,上线前应核对官方说明。
试点建议选低风险前台文案/样式类改动,强制要求截图断言与预发环境,禁止智能体直接推 main。细节以 Replit 产品公告为准。
对读者而言,可把本稿当作窗内一手动态的结构化摘要:关键数字、产品名与限定语均来自原帖或报道,未另行推断未披露的定价、配额或合规结论;若需上线决策,请回到厂商文档与控制台核对当前区域可用性。
Replit Cloud Agents for existing company codebases
Replit just opened its Cloud Agents to existing company codebases for all Pro and Enterprise customers. Anyone on a team can describe a product change in plain language and get back a tested pull request, with screenshots and video, for engineers to approve.
Quick context if you haven't used it. Replit is a browser-based tool where you describe software and an AI agent builds it. Until now that mostly meant new, standalone apps. This points the agent at the real product your company already runs.
Here's how it works. You describe the change, something like "fix the spacing on the billing page." The agent finds the relevant code, shows you a live preview of the running app, and tests the change in a copy of the app. It clicks through it, fixes what fails, and then opens a pull request in GitHub or GitLab. A pull request is the formal "please review and merge this" step engineers already use every day. An admin sets up the environment once, and you can even kick off a request by mentioning Replit in Slack.
Why this matters: it's the step past prototypes. Replit's own framing is that "changing your company's core product is still out of reach for anyone outside engineering." The designer, support lead, or ops person who actually knows what's broken can now hand engineering a working, tested change instead of a ticket that waits for a sprint.
And it doesn't go around engineers. They still review it and decide what ships. That's the right design. It moves Builders who use AI from side apps into the main product, without anyone pretending code review is optional.