plain text
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

LLMOps Tools: Chatbot or Workflow Builder

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)

Multi LLM Agent Collaboration Tools

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)

Browser-Use Agent Tools

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