RongQi · Unified Enterprise AI Resource Management Platform

Introduction¶

DeepSeek Harness (DSH) advocates the “everything is a plugin” architectural philosophy. Based on this architecture, 01men/ybkk-aios (Rongqi) has built an enterprise-level unified AI resource management platform. It aims to solve the problem of fragmented management across multiple resource types (IAM, models, applications, and skills). By leveraging a single identity, unified permissions, unified metering, and unified auditing, it enables unified access, scheduling, and control of AI resources.

Core Features¶

The platform primarily provides the following capabilities:
* Unified authentication and permissions: Supports IAM organizational structure, OIDC login, and scenario-level authorization.
* Resource onboarding and management: Includes a model access gateway, MCP service deployment and circuit breaking, and management of Agents and AI application entities.
* Skill Marketplace: Provides Skill submission, static scanning, version management, and marketplace listing workflow.
* Metering and auditing: Transparent usage metering, auditing and alerting, and internal cost drill-down.
* Workflow orchestration: Supports the workflow orchestration domain, including TF orchestration, state machines, and a template library.
* Model channel governance: Tiered routing, fallback chain, and budget circuit breaking.

Installation and Startup¶

Install the plugin¶

Install it via the DSH plugin command:

dsh plugin --profile web add github:01men/ybkk-AIOS

Local execution¶

After installation, install the dependencies and start the platform. On the first startup with an empty data directory, baseline initialization runs to generate the root organization and administrator account.

npm install
npm start

After startup, access http://127.0.0.1:7300 to enter the console.

Demo mode¶

If you need to quickly evaluate or test features, you can start demo mode. This mode automatically generates complete demo data and demo accounts.

DEMO_SEED=1 npm start

In demo mode, passwordless DingTalk login is available (Mock connector), and the passwords for all demo accounts are Ybk@2026.

Architecture and Configuration¶

Rongqi is itself a Cordis plugin tree. Each business domain maps to a plugin package that can independently declare its dependencies and permissions.

Dependencies and Storage Constraints¶

In development or standalone host mode, @deepseek-ai/cordis must exist only in devDependencies. If placed in dependencies, pnpm hoists it to the host, causing the plugin to resolve the wrong Cordis instance and breaking the service chain.

Platform service-key conflicts have been addressed:
* Storage key: Use opsStorage (not storage) to avoid conflicts with host-provided services.
* Tool key: tools is an intentional shared extension point. In standalone host mode, it is provided by ToolRuntimeLite; after being mounted into dsh, it becomes part of the native ToolRuntime.

Configuration Notes¶

• Cordis YAML: When checking out the source code for local development, the <PROJECT_ROOT> placeholder in cordis.yml must be replaced with an absolute path in file:/// format (bare drive-letter paths on Windows cause ESM errors).
• Same-instance dependency: For local development, node_modules/@deepseek-ai/cordis must point to vendor/cordis in the dsh source tree (Junction), ensuring that the plugin and host load the same instance.

Operations and Upgrade¶

Platform self-update¶

The platform supports automatic update checks and manual upgrades. The upgrade policy is determined by the administrator.
* Automatic check: Once every 24 hours by default; the first check occurs 15 seconds after startup. If a new version is found, a badge is displayed in the console.
* Manual upgrade: The console provides a one-click upgrade option, or use CLI commands.

    dshctl update status | check | apply [--dry-run]

Business Model and Compliance¶

• Pricing: Based on team + domain assets (e.g., industry knowledge graph packs, scenario packs). Manufacturing customers can use a private deployment/project-based model.
• M0 compliance: Externally, the platform only surfaces usage and internal cost drill-down. Usage-based billing, reselling, and settlement are discontinued. The wallet/settlement sub-systems have been taken offline and archived.
• M1-1 contract: Includes J1-J4 contracts and security hardening (e.g., security response headers enabled by default, default 730-day retention policy for usage events).
• IAW batch: Includes capabilities such as workflow orchestration, data element domain, model channel governance, Agent autonomy, and A2A.

Use Cases and Notes¶

• Use cases: Development and operations teams that need to manage AI models, MCP services, Agents, and skills within an enterprise in a unified way; enterprise environments that require transparent metering and auditing.
• Runtime permissions: The plugin runs with the permissions of the current DSH process. Before installing, verify the source code and license (MIT).
• References: Detailed design documents and contracts are available in the docs/ folder under the project directory.