Agent framework download trends

Daily downloads on the first vs the last date of a ~180-day window, for major agent-building libraries on PyPI and npm. Default bars use a 7-day average to remove weekday/weekend noise.

Generated 2026-05-19 13:02 UTC · PyPI data via pypistats.org · npm data via api.npmjs.org

PyPI — Python frameworks

Framework	Package	Daily downloads (7d avg)	vs 30d ago	vs 90d ago	Last ~180d total
LangChain	`langchain`	8,395,166	+7.5%	+15.1%	1,120,010,511
Pydantic AI	`pydantic-ai`	1,423,716	+91.3%	+226.3%	98,511,744
CrewAI	`crewai`	576,510	+148.3%	+261.7%	32,867,364
browser-use	`browser-use`	557,117	+47.8%	+354.9%	32,495,751
LlamaIndex	`llama-index`	466,465	+57.8%	+49.4%	52,965,667
AutoGen (agentchat)	`autogen-agentchat`	52,703	+42.8%	+100.7%	6,605,131
AutoGen (pyautogen)	`pyautogen`	29,256	-34.9%	+153.3%	3,421,307
smolagents	`smolagents`	23,884	+12.9%	+90.0%	2,596,569
fast-agent (mcp)	`fast-agent-mcp`	3,867	+36.3%	+293.6%	302,781

npm — JavaScript frameworks

Framework	Package	Daily downloads (7d avg)	vs 30d ago	vs 90d ago	Last ~180d total
Vercel AI SDK	`ai`	1,907,047	+16.5%	+79.4%	211,055,888
Mastra	`@mastra/core`	137,032	+39.4%	+119.4%	12,448,621

PyPI — first vs last date

log scale 7-day average (vs raw daily)

Log scale recommended — LangChain (~8.4M/day) dwarfs newer frameworks. With "7-day average" on, bars show the average of the first 7 days vs the last 7 days (smooths weekend dips).

PyPI — excluding LangChain, LlamaIndex, Pydantic AI

Focused view of newer/agent-native frameworks where the absolute numbers are more comparable.

npm — first vs last date

7-day average (vs raw daily)

npm package ai is the Vercel AI SDK; it's not exclusively used for agents (also chat UIs / generic LLM calls), so absolute volume overstates pure agent-framework usage.