A critique of Ollama uncovers its systematic downplaying of reliance on Georgi Gerganov's llama.cpp project alongside license violations and a 2025 custom backend that reintroduced resolved bugs. The primary source reports Ollama's README and marketing omitted llama.cpp references for over a year with binary distributions lacking required MIT copyright notices, as detailed in unresolved GitHub issue #3185 for 400+ days and partial acknowledgment only after PR #3700 (https://sleepingrobots.com/dreams/stop-using-ollama/, https://github.com/ollama/ollama/issues/3185). Ollama co-founders cited patching burdens before announcing transition from the 100k-star, 450-contributor upstream project. Post-fork implementation directly on ggml produced failures in structured output, vision models, GGML assertion crashes, and tensor support for models including GPT-OSS 20B, issues Gerganov publicly identified and that llama.cpp had addressed years earlier (https://github.com/ggerganov/llama.cpp). Original coverage missed Ollama's default unauthenticated localhost:11434 API creating local exploit surfaces for prompt injection, performance regressions versus native llama.cpp inference speeds, and maintenance lags in feature adoption, patterns synthesized from Ollama repository history, llama.cpp discussions, and Hacker News threads on VC influence diverging from local-first origins.