A new agent framework shows up every week, and most are already walking dead. Not badly written, just unmaintained six weeks from now. The skill is spotting which ones survive, and launch posts will not tell you: they are advertisements. - Our database watches 548 open source agent frameworks on six GitHub signals: stars, forks, contributors, maintenance cadence, issue response, release activity. - A repo with steady commits and a handful of real contributors is a different animal than a viral launch with three months of silence. - The data says consolidation: a few frameworks soak up almost all the activity. Good news. Your shortlist is short. - Data has a limit: frameworks cluster by philosophy. Graph control (LangGraph), role teams (CrewAI), conversational patterns (AutoGen). Match the philosophy to your head, then filter by activity. Want one built instead of researching? PrivateLLM deploy puts the private LLM underneath your agents on AWS for $50 plus usage. The framework database stays free.