ai-agent-platform

ai-agent-platform is an enterprise-grade AI Agent platform foundation designed for secondary development and production deployment, built on LangGraph / LangChain. The project adopts a core architecture decoupling the platform control plane from the Agent Runtime, separating platform governance (authentication, project management, audit, catalog management) from the Agent Runtime execution layer (graph orchestration, model assembly, Tools/MCP/Skills integration), enabling independent evolution and collaborative development.

The platform includes four default services: interaction-data-service (result domain data service), platform-api (platform backend control plane API), platform-web (platform frontend console), and runtime-service (LangGraph Agent Runtime), plus runtime-web as a debug frontend directly connecting to the Runtime. The frontend is built with Vue 3 + TypeScript + Vite, the backend with FastAPI, the Runtime with LangGraph / LangChain, and MCP tool integration with FastMCP.

The project provides a complete AI continuous programming Harness (boundaries, contracts, paradigms, closed-loops) that guides AI agents in continuous development, integration, and acceptance within defined boundaries. It also supports triggering AI agent automated environment deployment via natural language instructions. Starting with test engineering as the initial scenario, it has validated business use cases including AI intelligent review, AI-driven UI automation, Text-to-SQL, and multi-agent collaboration.

For deployment, it supports one-click local script startup and three Docker Compose modes (Runtime only, full stack without Nginx, full stack with Nginx), with model configuration supporting OpenAI-compatible APIs. The platform features a dual-path design — the platform path (platform-web → platform-api → runtime-service) and the debug path (runtime-web → runtime-service) run in parallel.

Architecture Layers:

┌─────────────────────────────────────────────┐
│  Control Plane                                │
│  ┌──────────┐  ┌──────────────┐             │
│  │platform-web│→│platform-api  │             │
│  └──────────┘  └──────────────┘             │
├─────────────────────────────────────────────┤
│  Agent Runtime Execution Layer               │
│  ┌──────────────┐  ┌─────────────────────┐  │
│  │runtime-service│  │runtime-web (debug)   │  │
│  └──────────────┘  └─────────────────────┘  │
├─────────────────────────────────────────────┤
│  Result Domain Data Service                  │
│  ┌───────────────────────────┐              │
│  │interaction-data-service    │              │
│  └───────────────────────────┘              │
└─────────────────────────────────────────────┘

Default Local Ports:

• interaction-data-service: 8081
• runtime-service: 8123
• platform-api: 2142
• platform-web: 3000
• runtime-web (optional): 3001

Quick Start:

scripts/dev-up.sh        # Start all services
scripts/check-health.sh  # Health check
scripts/dev-down.sh      # Stop

Unconfirmed Information: Open-source license type, specific Python version requirements, underlying database type for interaction-data-service, Windows compatibility, multimodal model configuration scope, authentication method details, horizontal scaling capability of each service.