Frameworks

Vibe Compiler

Self-compiling JavaScript tool that processes numbered markdown prompt stacks through an LLM to generate code, tests, and documentation, then uses that process to rewrite its own compiler. Works with OpenAI-compatible APIs and Claude via OpenRouter. MIT licensed, 67 GitHub stars.

LangChain

Open-source framework for building LLM applications, with integrations for over 100 model providers and a standard agent loop built around create_agent and middleware. Since its 1.0 release in October 2025, every LangChain agent runs on the LangGraph execution runtime.

CrewAI

Standalone Python framework, with no LangChain dependency, for multi-agent systems where agents get a role, a goal, and a set of tools, then collaborate in a crew on a shared task. Open-source core under MIT; its managed cloud platform offers a free tier plus a $25/month Professional plan, billed separately from LLM API costs.

AutoGen

Microsoft Research's early multi-agent framework for Python, placed into maintenance mode in October 2025 and now limited to bug fixes and documentation from community contributors. Still widely used for prototyping and teaching multi-agent patterns; Microsoft points new production projects to its successor, Microsoft Agent Framework.

LlamaIndex

Open-source data framework for retrieval-augmented generation and agents over enterprise data, with over 150 connectors through LlamaHub for sources like Notion, Slack, and SQL databases. Workflows, its event-driven orchestration layer, turns it from a pure RAG tool into a general agent framework, alongside newer products like LlamaParse for document understanding.

Parsel

Research framework that breaks a programming task into a hierarchy of functions described in natural language, then has a code LLM generate and test multiple implementations of each piece until the combination passes. Raised HumanEval pass rates from 67% to roughly 85% in the original paper.

LangGraph

LangChain's graph-based runtime for stateful agents, modeling each step as a node with conditional edges, explicit state, and checkpointing at every super-step for time-travel debugging. Surpassed CrewAI in GitHub stars in early 2026 and sees around 34.5 million monthly PyPI downloads.

DevOpsGPT

Open-source multi-agent system that turns natural-language requirements into working software, pulling and pushing code through Git. Works across any programming language and extends existing codebases, with around 6,000 GitHub stars.

StackSpot AI

Multi-agent platform covering the full software development lifecycle. Lets teams orchestrate specialized agents for code generation, review, test writing, and deployment, with governance across GitHub, GitLab, and Bitbucket.