What the signals say. The dominant pattern across freeintelligence's 24 signals is the rapid formalization of the "LLM Wiki" — the idea Karpathy kept predicting — into shipping standards and tools. Google's Open Knowledge Format (OKF) turns a directory of markdown files into a vendor-neutral way to feed curated context to foundation models. OpenAlex (a 20-person team rebuilt the Library of Alexandria, free) and MinerU (turn a 500-page contract into searchable text) are the ingestion side of the same thesis. The local-model movement is no longer fringe: Apple open-sourced coreai-models to put AI on 2 billion iPhones, Microsoft launched Frontier Tuning ("stop renting intelligence, own it"), and the loudest advice in the feed is "build llama.cpp from source and keep a local model in storage as an insurance policy."
Understand Anything is the closest existential threat: a free, viral codebase-KG that could expand to "understand anything." OpenAlex owns the open scholarly corpus. Firecrawl is consolidating the ingestion-on-demand layer. The window where "an AI that reads your stuff and answers questions" is a defensible product is narrowing fast — the open-source stack is assembling itself underneath it.
llama.cpp / Apple coreai. The strongest recurring signal in the feed is ownership of the model. A "download your intelligence, run it offline, no account" tier is both the insurance-policy framing customers respond to and the differentiator against the hosted incumbents.