AI 与无代码融合的开源平台，让编码 Agent 与人工协作搭建业务系统

NocoBase is an open-source TypeScript platform that merges AI with no-code building for business systems. It was created for teams that need CRM, ERP, admin dashboards, internal tools, or similar applications fast, but without trusting black-box generated code. Instead of asking AI to compose an entire application from scratch, NocoBase gives AI a real platform to stand on: a WYSIWYG interface, standard data models, permissions, workflows, and a plugin system. The result is an environment where AI accelerates delivery while the underlying infrastructure stays production-proven.

AI code generation is fast but often weak exactly where business systems are unforgiving: data models, field-level permissions, workflow state, and auditing. Recreating those from scratch on every project is risky and expensive. NocoBase treats them as prebuilt, production-tested infrastructure. AI agents and no-code users then assemble from these trusted modules. This narrows the gap between an AI-generated demo and a system that can actually run the business.

NocoBase deliberately keeps two building tracks. On the coding-agent track, tools such as Claude Code, Cursor, Codex, OpenCode, and TRAE get a CLI and can handle setup, development, migration, and release end to end. On the manual track, a WYSIWYG interface lets people switch between usage mode and configuration mode with one click, then visually review and adjust data models, pages, workflows, and permissions. Since both tracks work on the same system, the work can be split naturally: an agent scaffolds a model, a person polishes the UI and interactions, and the agent continues from there.

Beyond building the system, NocoBase puts AI to work inside it. Built-in AI employees receive business context and can handle analysis, Q&A, form filling, document recognition, risk monitoring, and task routing. Connected to workflows, they participate in decisions and execution on both front-end and back-end sides. External agents also connect through standard interfaces such as MCP, HTTP APIs, CLI, and skills. Platforms including OpenClaw, Hermes, Dify, Coze, and n8n can talk to the system, while channels like Telegram, WhatsApp, Slack, and Gmail can query data and trigger business workflows. Every AI employee has its own role and inherits the same granular permissions as human users, including field-level read/write control, with audit logs recording data changes and workflow triggers.

Reliability comes from built-in infrastructure rather than regenerated code. NocoBase ships data models, permission controls, workflows, and audit logs out of the box. Its data-model-driven design keeps business data in standard relational structures, decoupled from UI. You can use the platform's own database, external databases, or third-party APIs as data sources. The microkernel architecture treats everything as a plugin, so official and custom plugins can be mixed while staying aligned with the same system conventions. AI-generated plugins and manually built plugins follow the same rules, which also avoids vendor lock-in because business data remains in your own database.

A NocoBase app can be initialized from the official CLI. After installing the CLI, you can create a project like this:

npm install -g @nocobase/cli
nb --version
nb init --ui

nb init --ui bootstraps a full NocoBase application. If you want a coding agent to help you build on top of it, you can run the agent afterwards, for example:

codex # claude, opencode

Detailed instructions are available in the NocoBase docs section on installing an app and the AI Agent Integration Guide.

NocoBase is best suited to data-heavy business systems with structured relationships and permission requirements: CRM, ERP, project management, internal tools, salesforce-like platforms, and admin dashboards. It is also a good fit when you want to iterate with AI but keep human control over model, workflow, and access.

It may be too heavy for a one-off landing page or a simple prototype that does not need relational data, permissions, or audit. And while the AI coding agents perform real development, the safety guarantees depend on configuration: role permissions and audit settings must still be designed by someone. Teams adopting NocoBase should be prepared to think in terms of data models and plugins rather than "ask AI for a complete app and walk away".