Enterprise apps written by an AI agent on ObjectStack, each one verified before it is shown.
Every app here starts from a scenario card: a business requirement plus the checks a correct app must pass. An AI coding agent builds the app from the requirement alone. Code, not a person, then checks the result:
- Structure: the objects, relationships and navigation the requirement implies.
- Data questions (问数): 8–10 business questions per app, answered by an agent connected over MCP. Each answer is compared with a truth computed from the seed data. The same question is asked as different roles, so a role must never see what it is not allowed to see.
- Write operations: entries, edits and workflow steps the agent performs for a user. Irreversible actions must wait for a human to confirm. An overstep must be refused, and a write that breaks a validation rule must be rejected.
- One change request: a one-sentence customization, applied and re-checked.
Only apps that pass appear here. The whole gallery is rebuilt on every ObjectStack release, so nothing shown is pinned to an old version.
Status: empty. The first verified apps arrive when the scenario pipeline lands (tracked in objectstack-ai/objectstack#30349).
| Path | What it is |
|---|---|
apps/<scenario>/ |
One verified app: its metadata source, run record and scorecard, screenshots, and the data-question transcript for each role. |
index.json |
A machine-readable index of every app: function, title, screenshots, question count, framework version, run date. Landing pages on objectos.ai are built from it. |
The apps are general business functions, not industry verticals: sales, service, procurement and supply chain, finance, HR, admin, projects and legal.
- Read it. Each app's metadata is plain TypeScript and JSON. It is small enough to review as a diff, which is the point: the AI writes it, and you review it and sign off.
- Run it locally. Start a project with
npm create objectstackand copy the app's metadata in. Runos validateandos verify, then start it withos dev. - Try it in ObjectOS cloud. A one-click trial is planned.
- Not hand-maintained. Every file under
apps/andindex.jsonis generated, and a hand edit is overwritten on the next release. - Not a template catalogue with an upgrade promise. A starter template that graduates to the ObjectStack marketplace installs unmanaged: the copy becomes yours.
Apache-2.0. See LICENSE.
由 AI 在 ObjectStack 上写出的企业应用,每一个都先验证、再展示。
每个应用都从一张场景卡开始:一段业务需求,加上一个正确的应用必须通过的检查。AI 编程助手只凭需求把应用搭出来,然后由代码(不是人)检查结果:
- 结构:需求所隐含的对象、关联和导航。
- 问数:每个应用 8–10 个业务问题,由通过 MCP 接入的 agent 回答。每个答案都和根据种子数据算出的标准答案比对。同一个问题会换不同角色再问,确保角色看不到无权看的数据。
- 写入操作:agent 代用户做的录入、修改和流程操作。不可逆的操作必须等人确认,越权的操作必须被拒绝,违反校验的写入必须被拦下。
- 一句话改需求:做一次小改动,改完再验一遍。
只有全部通过的应用才会出现在这里。每次 ObjectStack 发版,整个画廊都会重新生成,所以展示的内容不会停留在旧版本上。
现状: 目前为空。场景产线落地后,第一批验证通过的应用就会出现在这里(进度见 objectstack-ai/objectstack#30349)。
覆盖的是通用职能场景,暂不按行业细分:销售、客服、采购与供应链、财务、人事、行政、项目、法务。
本仓库的内容全部由产线生成,手工修改会在下次发版时被覆盖。毕业到应用市场的入门模板以非受管方式安装:装上之后,那份应用归你所有。