# 5 AI-agent open-source projects worth onboarding

All links fetched and verified live on 2026-10-24 (HTTP 200). Star counts from the GitHub API at fetch time.

| # | Project | Stack / license | One-line rationale |
|---|---|---|---|
| 1 | **LangGraph** — https://github.com/langchain-ai/langgraph | Python/JS, MIT, ~43.8k ★ | Graph-based orchestration with durable state and human-in-the-loop checkpoints — the safest bet when an agent needs to survive restarts and branching. |
| 2 | **CrewAI** — https://github.com/crewAIInc/crewAI | Python, MIT, ~60.0k ★ | Role-based multi-agent crews with a clean task/handoff model, so teams of agents are described declaratively instead of glued together in code. |
| 3 | **Pydantic AI** — https://github.com/pydantic/pydantic-ai | Python, MIT, ~20.8k ★ | Brings Pydantic's type contracts to agent I/O, making tool schemas and structured outputs explicit and validated rather than prompt-hoped. |
| 4 | **smolagents** — https://github.com/huggingface/smolagents | Python, Apache-2.0, ~30.0k ★ | A ~barebones "agents that think in code" library; tiny surface area and model-agnostic, ideal for embedding an agent in an existing app. |
| 5 | **OpenAI Agents SDK** — https://github.com/openai/openai-agents-python | Python, MIT, ~30.3k ★ | Lightweight multi-agent primitives (agents, handoffs, guardrails, tracing) with first-class tool calling — the lowest-friction starting point for an OpenAI-centric stack. |

**How they differ in one breath:** LangGraph and CrewAI are *orchestration* frameworks (graph vs. roles); Pydantic AI and smolagents are *agent libraries* (typed vs. minimal); the OpenAI Agents SDK is a *vendor-native toolkit* optimized for a single provider's ecosystem.

**Onboarding tip:** all five ship a pip-only install and a runnable example in their README, so a newcomer can go from clone to a working tool-calling agent without external services. Every repository URL above returned HTTP 200 at fetch time.