TypeScript framework for durable AI workflows and agents. Long-running tasks with automatic retries, checkpointing, and observability. Boundary case: also runs as a hosted platform, but the code is the entry point.

Role-playing multi-agent framework in Python. Define agents with roles, goals, and tools; assemble them into crews that collaborate on tasks. Widely adopted as the default "let me try multi-agent" starting point.

Graph-based agent framework from LangChain. Build stateful multi-actor applications as directed graphs where nodes are agents or tools and edges encode control flow. The default for anything that needs branching or human-in-the-loop.

OpenAI's official Python SDK for agents. Ships handoffs, guardrails, tracing, and a ReAct-style loop built around the Responses API. The reference for anyone starting on OpenAI models.

Type-safe agent framework from the Pydantic team. Structured outputs, dependency injection, and model-agnostic tool calls with Python type hints as the contract. First-class TypedDict validation of model output.

TypeScript-first agent framework. Agents, workflows, tools, and evals in one library, aimed at devs building agents inside Next.js/Node apps rather than Python notebooks.