地端 AI Agent + AutoGen
Token Intelligence無極限
What and Why is Agentic AI?
In computer science, a software “agent” is a computer program that acts for a user or another program in a relationship of agency.
- Agentic AI is an advanced form of artificial intelligence that possesses autonomous decision-making capabilities, allowing it to perform complex tasks and interact with its environment without constant human intervention.
Local LLM AI Agent benefits
- Less latency(faster response)
- Enhanced privacy
- Cost efficiency
- Offline functionality
- Other reasons
AutoGen for .NET Overview
AutoGen 
An open-source framework specific for build LLM Agentic AI software, provide both Python & .NET C#:
- Python: https://microsoft.github.io/autogen/stable/
- .NET C# porting:
Important
It will also as Semnantic Kernel, being replaced by Microsoft Agent Framework in near future:
https://devblogs.microsoft.com/foundry/introducing-microsoft-agent-framework-the-open-source-engine-for-agentic-ai-apps/
And because of that, another related open source project AG2: Open-Source AgentOS for AI Agents is started to provide a truly “Open-Source” Agentic AI framework.
AutoGen design concepts
AutoGen (Python) is designed to simplify the development of agentic AI applications by providing:
Agent:
Autonomous entities that can perform tasks, make decisions, and interact with other agents or users.Agent Runtime Environment:
Standalone or Distributed runtime environment, facilitates communication between agents, manages their identities and lifecycles, and enforce security and privacy boundaries.![]()
So it is possible to run agents across multiple machines:
![]()
Memory:
Included in “AgentChat” package, Mechanisms for agents to retain and recall information, facilitating long-term interactions and learning.Messaging:
Provids Topic and Subscription ways for either direct or broadcast communication between agents.Tool Use:
Enables agents to utilize external tools and APIs to enhance their capabilities and perform complex tasks.Workbench:
Provides a collection of Tools for agents to use, including Model Context Protocol (MCP).Logging & Monitoring:
Built-in support for logging and monitoring agent activities via OpenTelemetry, aiding in debugging and performance analysis.Group Chat/Workflow:
Supports multi-agent various types of group chat and workflow orchestration, enabling complex interactions and collaborations among agents.
Local LLM architecture overview
We only care about using Local LLM model that can provide OpenAI compatible RESTful API endpoint, so that AutoGen for .NET agent can use it as LLM backend service.
xPU TOPs myth
Currrently most of NPU(s) are only support ONNX model format, and lack of some advanced features(ex: function-calling) that Agentic AI usually need.
Total VRAM myth
Total VRAM myth
ASUS Ascent GX10
It has extensible cable that can connect multiple machines together for settup up local AI inference cluster.
External Thunderbolt GPU enclosure
Current speed leader (nVidia CUDA wins)
The more you buy, the more (Time) you save!
Demo
Q & A
Any Questions?
🙋♂️🙋♀️🙋
