Every notable framework for building AI agents, with live GitHub stars, npm and PyPI downloads, language, license, and the trade-offs nobody puts on the homepage.
| Maintainer | Language | License | Compare | |||||
|---|---|---|---|---|---|---|---|---|
| LangChain Inc. | Mixed | MIT | 138.6K | 2.4M | 299.7M | Today | ||
| Microsoft Research | Python | MIT | 58.7K | — | 1.5M | 2 mo ago | ||
| CrewAI Inc. | Python | MIT | 52.9K | — | 14.8M | Yesterday | ||
| LlamaIndex Inc. | Mixed | MIT | 49.9K | — | 12.3M | 1 wk ago | ||
| Agno Inc. | Python | apache-2 | 40.5K | — | 4.4M | Yesterday | ||
| LangChain Inc. | Python | MIT | 33.6K | — | 54.5M | 5 days ago | ||
| Cursor (Anysphere) | Mixed | Proprietary | 32.9K | 212.1K | 64.9K | 3 wk ago | ||
| Hugging Face | Python | apache-2 | 27.7K | — | 694.5K | 4 days ago | ||
| OpenAI | Python | MIT | 26.9K | — | 31.7M | Yesterday | ||
| deepset | Python | apache-2 | 25.5K | — | 934.0K | Yesterday | ||
| Mastra AI | TypeScript | apache-2 | 24.8K | 965.7K | — | Today | ||
| Vercel | TypeScript | apache-2 | 24.7K | 14.2M | — | Yesterday | ||
| Python | apache-2 | 20.0K | — | 24.4M | 2 days ago | |||
| Pydantic | Python | MIT | 17.6K | — | 38.6M | Yesterday | ||
| Microsoft | Mixed | MIT | 11.1K | — | 1.1M | Yesterday |
An agent framework is a software library that turns a language model into something that can take actions, calling tools, holding memory, and looping in a human when it is unsure.
Picking one shapes hiring, infrastructure, and how fast you can ship. The smallest framework that covers your real requirements almost always beats the most popular one.
Filter by the language your team already writes in, then by license if you have constraints, then by the capabilities you actually need. Pin two or three to the comparator and read the trade-offs side by side before you commit.
Side-by-Side
Pick any two or three and see stars, downloads, capabilities, and the case for each one in a single matrix you can share.
Run the Numbers
Estimate per-task and monthly cost across LangGraph, CrewAI, Mastra, and the OpenAI Agents SDK on the same workload. Workflow-aware, with framework overhead baked in.
An agent framework is a software library that lets you turn a language model into something that can take actions. It handles the loop of prompting the model, calling tools or APIs, remembering past steps, and asking a human when it is unsure. Without one, every team rebuilds the same plumbing.
Open source wins on cost, portability, and how deep your team can go on tuning. Hosted frameworks like LangSmith or LlamaIndex Cloud win on time to first demo and on built-in tracing. Most production teams end up self-hosting an open framework and adding a hosted observability tool on top.
LangGraph, CrewAI, and AutoGen are the three teams ship multi-agent systems on most often. LangGraph gives you the most control with a typed state graph. CrewAI is the fastest to prototype with role-based agents. AutoGen is strongest for research-style multi-turn conversation between agents. The comparator on this page lays them side by side.
GitHub stars, forks, contributors, last commit, npm weekly downloads, and PyPI monthly downloads come from a scheduled sync against the source registries. Capability flags like multi-agent, streaming, tool use, and human-in-the-loop are editorial, set by reading the docs and the source code. Trend arrows show change since the last sync.
Adopt one. The frameworks listed here have hundreds of contributors and proven patterns for tool use, retries, streaming, and tracing. Building your own makes sense only if you have a constraint that no open framework supports, and even then, fork an existing one before starting from scratch.
LangChain is the general-purpose toolkit for chaining LLM calls and tools. LangGraph is its newer sibling for stateful, multi-step workflows with branching. CrewAI is a Python framework focused on role-based multi-agent crews. AutoGen comes from Microsoft Research and specializes in conversational multi-agent setups. They overlap but have different sweet spots, which the comparator makes explicit.
Stars, downloads, and commit timestamps refresh on a daily cron from GitHub, npm, and PyPI. Capability flags are reviewed monthly. When a framework ships a major release or gets archived, this directory updates within a day.
