DevOps (개발·운영 통합)
│
├── 공통 핵심: CI/CD, 버전 관리, 모니터링, 자동화
│
└─▶ MLOps (Machine Learning Operations)
│
├── 특징: 데이터 버전 관리, 모델 학습/배포 파이프라인, 파라미터 튜닝, Drift 감지
│
└─▶ LLMOps (Large Language Model Operations)
│
├── 특징: Prompt 관리, Prompt 버전 관리, A/B 실험, 사용자 피드백 루프
├── LLM API 연결, Cost 최적화, Latency 관
├─ 오픈소스: Agenta, LangChain, PromptLayer
└─ 상용: Vertex AI (구글), OpenAI Playground, Anthropic Console
DevOps / MLOps / LLMOps 핵심 개념 및 진화 단계
| 계층 | 핵심 개념 | 설명 |
| DevOps | Software Development Lifecycle Automation | Development와 Operations를 통합하여 소프트웨어 개발 및 배포 과정을 자동화하고 효율성을 높이는 방법론입니다. CI/CD, Version Control, Monitoring, Automation이 핵심입니다. |
| MLOps | Data/Model-Centric DevOps | Machine Learning 모델의 개발, 배포, 운영을 자동화하고 관리하는 방법론입니다. DevOps의 원칙을 따르면서 Data Versioning, Model Training/Deployment Pipeline, Parameter Tuning, Data/Model Drift Detection과 같은 Machine Learning 특유의 요구사항을 다룹니다. |
| LLMOps | LLM-Specific MLOps | Large Language Model (LLM)의 개발, 배포, 운영을 위한 MLOps의 진화된 형태입니다. LLM의 특성을 고려하여 Prompt Management, Prompt Versioning, A/B Testing, User Feedback Loop를 다룹니다. 또한 LLM API Integration, Cost Optimization, Latency Management가 중요하게 다뤄집니다. |
| Agentic AI | Autonomous Decision and Action System | 단일 LLM의 한계를 넘어, Reasoning, Planning, Tool Use, Self-Correction을 통해 복잡한 작업을 자율적으로 수행하는 AI 시스템입니다. 특히 Mixture of Agents, Role-based Collaboration, Graph-based Computation, Peer Review 구조 등을 통해 Hallucination 현상, 신뢰성, Context 유지 문제를 개선하려는 시도가 활발합니다. |
DevOps / MLOps / LLMOps 대표 도구
| 계층 | 핵심 개념 | 대표 Open Source | 대표 Commercial |
| DevOps | Software Development Automation, Build/Deployment, Monitoring | Jenkins, ArgoCD, GitLab CI | AWS CodePipeline, GitHub Actions |
| MLOps | Data Management, Model Training/Tuning/Deployment, Monitoring | MLflow, Kubeflow, DVC | Google Vertex AI, AWS SageMaker |
| LLMOps | Prompt Management, A/B Testing, User Feedback | Agenta, LangChain, PromptLayer, OpenPipe | Google Vertex AI (LLM 부분), OpenAI Playground, Anthropic Console |
LLM을 빠르게 배포하고 싶은 사용자를 위한 경량 Chatbot/Workflow Builder 도구들
| 이름 | 특징 | Open Source/Commercial | 가격 정책 |
| LangChain | Chain/Flow Composition for LLM, Agent Management, Tool Integration | Open Source | Free (Self-host) |
| Flowise | GUI (LangChain-based), Drag & Drop Pipeline, LLM Workflow Builder | Open Source | Self-host Free / Cloud: $35/month |
| Agenta | Prompt Versioning, Parameter Tuning, A/B Experimentation, Feedback Loop Support | Open Source | Free (Self-host) |
| PromptLayer | LLM Call Logging, Prompt Execution Tracking, Versioning | Open Source (Core SDK) | SaaS Paid Plans Available |
| OpenPipe | Automated Fine-tune Dataset based on LLM Call Logs | Open Source | SaaS Paid Plans Available |
| Google Vertex AI (LLM Portion) | GUI, Gemini, Prompt Management, A/B Experimentation, Pipeline Integration | Commercial | Usage-based Billing |
| Anthropic Console | GUI, Claude Prompt Experimentation, Version Comparison | Commercial | Usage-based Billing |
| OpenAI Playground | GUI, GPT Prompt Playground, Parameter Experimentation | Commercial | Usage-based Billing |
| Google Gemini CLI | CLI, Gemini 2.5 Pro Model-based, MCP Support, Node.js-based | Open Source | Free for Personal Accounts (60 model requests/minute, 1,000 model requests/day) |
| Claude Code | CLI, Claude 4 Model-based, MCP Support | Commercial | Monthly $20 Pro Plan Subscription Required |
| GPTme | CLI-based Custom LLM Chatbot Builder | Open Source | Free (Self-host) |
단일 LLM의 한계 (Hallucination, Reliability, Context Maintenance)를 극복하기 위해 Role-based Collaboration, Graph-based Computation, Peer Review 구조 등을 활용하는 Agent Collaboration 도구들
| 도구명 | 주요 특징 | Source Code (License) | 가격 정책 |
| LangGraph | LangChain-family Graph Agent, Supports Cyclic/Branching Structures, LangSmith Integration | Open Source (MIT) | Self-host Free / Platform: $39/month (LangSmith Plus), Per-node Execution Billing |
| Camel-AI | LLM Role-play Q&A, First Public Multi-Agent Experimentation Framework | Open Source (Apache 2.0) | 100% Free (Uses User's API Key) |
| CrewAI | Role-based Collaboration Engine, Provides No-code UI Studio | Open Source (MIT) | Self-host Free / Paid Plans (Basic: $99/month, Standard: $500/month, Enterprise: $1,000+/month) |
| MetaGPT | Mixture-of-Agents Structure, Virtual Software Company Simulation | Open Source (MIT) | Free (API Call Costs Extra: ~$0.2/example) |
| AutoGen | Microsoft Research, Event-driven Multi-Agent, Various LLM Integrations, Community Expansion for Education/Research | Open Source (MIT) | Free (Uses User's API Key) |
LLM을 단순히 API로 호출하는 한계를 넘어, 웹 브라우저를 직접 조작하여 'Click / Scrape / Form Filling / Posting / Crawling' 등을 사람처럼 수행하는 Agent.
| 도구명 | 주요 특징 | Source Code/Type | 가격 정책 |
| Proxy Lite (Convergence AI) | Web Automation Generative AI Proxy | Closed Source | Service Discontinued (Acquired by Salesforce in June 2025). |
| Nanobrowser | Chrome extension for AI-powered web automation, Local Execution, Various LLM Integrations | Open Source (Apache 2.0) | Completely Free (Uses Personal API Key) |
| Browser-Use | Python-based Browser Automation Framework, Multi-tab, Visual+HTML Extraction, Self-recovery, No-code Navigator SaaS Provided | Open Source (MIT) | Self-host Free / API Access: Pay-as-you-go / Navigator: $200/month |
| Browserbase | Large-scale Headless Browser Infrastructure, Playwright/Puppeteer/Selenium Compatible, Stealth Mode, Proxy Support | Closed Source | Free: (1 hour/month) / Developer: $20/month / Startup: $99/month |
| Prompteus | No-code Prompt & Workflow SaaS, Browser Agent Beta (2024~), Semantic Caching, Multi LLM Support | Closed Source | $5 / 100K Requests |