HN 표시: AgentSwarms – 에이전트 AI를 배울 수 있는 무료 실습 놀이터, 설정 없음

Don't just read about agents. Build them. AgentSwarms is the interactive school for Agentic AI. Five tracks, forty-plus lessons, thirty-plus runnable agents — from your first prompt to multi-agent swarms in production. Six lessons. From "what's an agent?" to "I shipped a swarm." Every lesson is interactive. You read a concept, then run a live agent that demonstrates it — prompts and all. Prompts & System Messages Learn how an agent's personality, role, and constraints are shaped by the system prompt. See the same model behave like a teacher, a lawyer, or a sarcastic pirate — just by changing words. - Anatomy of a great system prompt - Few-shot vs zero-shot patterns - How temperature changes creativity vs accuracy RAG & Knowledge Bases Watch a generic chatbot transform into a domain expert by grounding its answers in your documents. Real citations, real docs, no hallucinations. - Why retrieval beats fine-tuning for facts - Chunking, embeddings, and citations - When RAG fails — and how to detect it Tools & Function Calling Give your agent superpowers. Connect it to APIs, MCP servers, and webhooks so it can actually do things — fetch data, send emails, run SQL. - OpenAI tool-call schema, plain English - MCP servers in 5 minutes - Designing safe, idempotent tools Guardrails & HITL Production agents need brakes. Add input/output filters, PII detection, content safety, and human-in-the-loop approvals for risky actions. - PII redaction and prompt-injection defense - Approval inboxes for high-risk actions - Cost & rate-limit guardrails Multi-Agent Swarms One agent is a worker. A swarm is a team. Build researcher → writer → reviewer pipelines with explicit handoffs and shared memory. - Orchestrator vs peer-to-peer patterns - Routing and handoff messages - When to split an agent into a swarm Observability & Evals If you can't trace it, you can't trust it. Inspect every token, tool call, and dollar spent — and learn how to evaluate agent quality systematically. - Reading execution traces like a pro - Token, latency & cost dashboards - Building your first eval suite Learn by doing, in four steps No installs. No API keys to start. Open a demo, follow the guided prompts, then make it your own. Try a Live Demo Start with the Templates gallery. Click any template — Product Support, Research Assistant, Code Reviewer — and a fully working agent is provisioned for you in seconds. Follow the Guided Tour Each demo opens in the Playground with a side-panel lesson. Suggested prompts walk you through RAG, guardrails, and approvals one checkpoint at a time. Fork & Experiment Tweak the system prompt, swap models (AgentSwarms AI, OpenAI, Gemini, Grok, Claude…), wire up your own knowledge base. Break things — that's how you learn. Build Your Own Apply what you learned. Compose your own agents, chain them into a swarm, and watch your traces light up in the observability dashboard. The Agentic AI vocabulary, demystified Every term you'll hear in agent papers, blog posts, and Twitter threads — explained in one line. - Agent - An LLM with a system prompt, optional tools, and memory — capable of multi-step reasoning toward a goal. - RAG - Retrieval-Augmented Generation. Inject relevant chunks from your docs into the prompt so the model can cite real sources. - Tool / Function call - A typed action the model can invoke (search_web, send_email, query_db). The agent decides when to call it. - Guardrail - Rules that filter input or output — PII redaction, profanity blocks, schema validation, cost caps. - HITL - Human-in-the-Loop. The agent pauses for human approval before doing something risky (refunds, deletes, sends). - MCP - Model Context Protocol. A standard way to expose tools and data sources to any compatible agent. - Swarm - Multiple specialized agents that hand off work to each other — researcher → writer → reviewer. - Eval - A test suite for agents. Score outputs on accuracy, format, safety, cost — not just vibes.