Auto-sync
This commit is contained in:
@@ -1,42 +0,0 @@
|
||||
---
|
||||
id: agent
|
||||
title: "Agent"
|
||||
type: concept
|
||||
tags: [AI, autonomous, tool-use]
|
||||
sources:
|
||||
- "[[LLM Terms Framework]]"
|
||||
last_updated: 2025-12-20
|
||||
---
|
||||
|
||||
## Definition
|
||||
|
||||
Agent(智能体)是LLM+MCP的组合,LLM负责给出步骤,MCP负责实际执行。
|
||||
|
||||
## How It Works
|
||||
|
||||
1. LLM理解用户意图
|
||||
2. LLM规划执行步骤
|
||||
3. MCP调用外部工具执行
|
||||
4. 结果反馈给LLM
|
||||
5. LLM继续下一步或返回结果
|
||||
|
||||
## Key Capabilities
|
||||
|
||||
- 自主决策
|
||||
- 工具调用
|
||||
- 任务分解
|
||||
- 迭代优化
|
||||
|
||||
## vs Vanilla LLM
|
||||
|
||||
| 维度 | Vanilla LLM | Agent |
|
||||
|------|-------------|-------|
|
||||
| 能力 | 仅生成文本 | 执行实际操作 |
|
||||
| 工具调用 | 无 | 有 |
|
||||
| 自主性 | 低 | 高 |
|
||||
| 幻觉风险 | 高 | 低(可验证) |
|
||||
|
||||
## Connections
|
||||
- [[Agent]] ← combines ← [[LLM]] + [[MCP]]
|
||||
- [[Agent]] ← extends ← [[LLM]]
|
||||
- [[Agent]] ← uses ← [[工具调用]]
|
||||
Reference in New Issue
Block a user