Introduction
AgentBuilder is an industrial-grade AI Agent development platform independently developed by Advantech, purpose-built for manufacturing enterprises. Centered on the three core pillars of data, tools, and knowledge, the platform delivers an MCP service catalog and visual agent-orchestration capabilities. By establishing a unified industrial data warehouse, metrics-driven insights, and an enterprise private knowledge base—and by integrating MCP tools purpose-built for equipment O&M, data analytics, and knowledge-management scenarios—it empowers manufacturers to rapidly build and deploy typical AI use cases such as predictive equipment maintenance and intelligent production-KPI analysis.
Features
- 01
Unified Data Foundation
Leveraging Advantech’s IoTSuite industrial IoT platform to ingest device data and unify integration with IT systems (DataInsight) and enterprise knowledge (KB Insight), breaking down data silos and providing AI with complete contextual information.
- 02
MCP Standardized Tool Ecosystem
Provides an enterprise-grade private MCP service catalog, compatible with both in-house developed and third-party tools, enabling agents to support 'plug-and-play' functionality. It includes built-in standardized MCP Servers for device monitoring, metric querying, knowledge retrieval, and more. Agents can securely invoke these services like function calls, seamlessly bridging thought and action.
- 03
Three-Stage Agent Orchestration
Supports no-code (via prompts), low-code (using open-source workflow platforms like Dify), and full-code (with AgentHub + LangGraph), covering the full spectrum of use cases—from rapid prototyping to multi-agent collaboration.
- 04
Out-of-the-Box Pre-Built Agents
Provides three ready-to-use agents for device operations & maintenance, data analysis, and knowledge-based Q&A, with support for rapid extension into manufacturing-specific domains such as production, energy, and supply chain.
- 05
Flexible Large Model Integration
Supports integration with public cloud models such as Azure OpenAI and Qwen.Enables on-premises private deployment and management of LLMs (Large Language Models), embedding models, and reranker models, exposing them via standardized APIs through Xinference or Ollama to ensure data security.
- 06
Full-Stack On-Premises Deployment
The AgentBuilder platform can be deployed within an enterprise’s internal network or private cloud. It supports a decoupled “platform + inference” dual-node architecture, separating the orchestration platform from the large model inference servers. This architecture helps ensure that data remains within the enterprise boundary and models stay on-premises, supporting manufacturing enterprises’ needs for security, compliance, and high availability.
Demonstration
Metric Insights
Unify business and technical definitions, break down data silos, and establish an authoritative metric system to ensure consistency and trustworthiness across the entire chain—from metric definition, calculation, to application.
Data Analysis Agent
Based on a unified semantic layer, bridge the gap from "seeing numbers" to "understanding insights." Deliver explainable and traceable analysis results through intelligent data querying and root-cause attribution.
MCP Catalog
MCP Catalog offering detailed descriptions, tool listings, and an online Playground—accelerating AI agent enablement.
MCP Server
Pre-configured standard MCP Servers managed uniformly via MCP Gateway—enabling seamless tool extension for AI agents.
Knowledge Management
The intelligent knowledge management platform builds AI-Ready Knowledge, transforming knowledge into conversational, reasoning-capable, and iteratively evolving intelligent assets.
Agent Orchestration
An Agent orchestration engine that automatically coordinates LLMs, MCP tools, and knowledge resources to enable multi-step reasoning, conditional logic, and task closure.
Product Update
- Version
- 2.0.0
- Release Date
- 2025/12/02
- What's new in this release?
- AgentBuilder 2.0.0 is now officially released, featuring a dedicated model management module that is fully independent and has no dependency on Dify.
- Language Supported
- English (US), 繁體中文, 简体中文
Additional Information
- Published by
- Advantech
- Category
- Copyright
- Advantech