分類 聊天機器人 生成圖片 生成影片 音樂/語音 寫程式 寫作 生產力 搜尋/研究 AI 代理 免費額度總表

Getting Started - Build Your First AI Agent Team

Getting Started with AMPM-AIOPS

Build your first autonomous AI agent team in 10 minutes.


Prerequisites


Step 1: Installation

# Clone the repository
git clone https://github.com/ampm-aiops/ampm-os.git
cd ampm-os

# Install dependencies
pip install -r requirements.txt

# Configure environment
cp .env.example .env
# Edit .env with your API keys

Step 2: Define Your Agents

Create agents.yaml:

agents:
  boss:
    name: "AMPM Boss"
    level: 5
    role: "orchestrator"
    model: "gpt-4"
    
  qiyuan:
    name: "Qiyuan (Research)"
    level: 2
    role: "researcher"
    model: "gpt-3.5-turbo"
    tools:
      - web_search
      - read_file
      
  jueze:
    name: "Jueze (Execute)"
    level: 4
    role: "executor"
    model: "gpt-4"
    tools:
      - write_file
      - execute_code
      - deploy

Step 3: Create Boss Agent

from ampm import BossAgent, AgentRegistry

# Load agent registry
registry = AgentRegistry("agents.yaml")

# Create boss agent
boss = BossAgent(
    registry=registry,
    model="gpt-4"
)

# Receive user goal
goal = "Improve website SEO and increase traffic by 50% in 3 months"
plan = boss.receive_goal(goal)

print(f"Plan: {plan}")

Step 4: Execute the Plan

# Boss automatically decomposes and assigns tasks
boss.execute()

# Monitor progress
while not boss.is_complete():
    status = boss.get_status()
    print(f"Progress: {status}")
    time.sleep(10)

# Get final report
report = boss.get_report()
print(report)

Step 5: Verify Results

# Verify agent checks completion
verifier = registry.get_agent("verifier")
verification = verifier.verify(boss.get_results())

if verification.passed:
    print("✅ All tasks completed successfully!")
else:
    print(f"❌ Issues found: {verification.issues}")

Agent Registry Reference

AMPM-AIOPS Agent Registry

Version: 0.1 (Initial)
Last Updated: 2026-07-28
Maintainer: 黑曜 (HeiYao)


Architecture Overview

                 AMPM Boss
              (總控 Agent / CEO)
                    |
        --------------------------
        |            |           |
      啟淵         覺澤        黑曜
   (Agent A)    (Agent B)   (協調/管理)

Agent Specifications

1. AMPM Boss

AttributeValue
Role總控 / Orchestrator
ModelTBD
System PromptTBD
Capabilities接收目標 → 分派任務 → 管理工作流 → 整合結果
ToolsAll
MemoryFull (短期 + 長期 + 知識庫)
Status🚧 Planning

Description:
類似公司的 CEO / 專案經理。負責理解 Howard 的目標,拆解成任務,分配給啟淵/覺澤,並監控執行品質。


2. 啟淵 (Qiyuan)

AttributeValue
Role研究主管 / Research Lead
Modelglm-5.2 (Cloudflare Workers AI)
System PromptDefault Hermes (❌ 需自訂)
Capabilities分析問題 → 搜尋資料 → 規劃策略 → 提出方案
Toolsweb_search, web_extract, read_file, session_search
MemoryResearch-focused
Status⚠️ Needs Role Definition

Current State:

Planned Role:


3. 覺澤 (Jueze)

AttributeValue
Role工程主管 / Engineering Lead
Modelglm-5.2 (Cloudflare Workers AI)
System PromptDefault Hermes (❌ 需自訂)
Capabilities修改文件 → 執行任務 → 操作工具 → 測試結果
Toolsterminal, patch, write_file, read_file
MemoryProject-focused
Status⚠️ Needs Role Definition

Current State:

Planned Role:


4. 黑曜 (HeiYao)

AttributeValue
Role協調者 / Coordinator
ModelTBD
System Prompt見下方
Capabilities任務協調 → 進度追蹤 → 品質把關 → Howard 溝通
Toolsdelegate_task, cron, memory, session_search
MemoryFull
Status🟡 In Progress

Description:
負責協調啟淵和覺澤的工作,確保任務按時完成,並向 Howard 回報進度。不直接執行開發或研究,而是管理和監控。


Next Steps

Phase 0: Agent Definition (當前)

Phase 1: Communication Protocol

Phase 2: Memory Sharing


Communication Protocol (Draft)

Task Assignment Format

task_id: AMP-001
assigned_to: qiyuan
priority: P0
objective: "研究競品 LangGraph 的優缺點"
deliverables:
  - type: report
    format: markdown
    path: ~/ampm-os/reports/langgraph-analysis.md
deadline: 2026-07-30
status: pending

Status Update Format

task_id: AMP-001
agent: qiyuan
status: in_progress
progress: 60%
blockers: []
eta: 2026-07-29

Notes

  1. 啟淵/覺澤目前是「空白」
    他們的角色定義需要明確寫入各自的 SOUL.md,而不是依賴預設 prompt。

  2. 黑曜的角色定位
    不是「第三個 Agent」,而是「協調層」。未來可能被 AMPM Boss 取代。

  3. 模型選擇
    目前都用 glm-5.2(免費),未來可能需要根據任務複雜度切換模型。

  4. 記憶共享
    目前三個 Agent 的記憶是隔離的,未來需要建立共享機制。