April 16, 2026
Drama llama strikes again
Stop Using Ollama
Fans turn on Ollama: “Credit llama, fix bugs, stop the spin”
TLDR: A blowup claims Ollama leaned on llama.cpp without clear credit, then switched engines and sparked new bugs; now users are split. Critics slam licensing missteps and broken downloads, while fans defend the one‑command convenience and model “hotswapping,” making trust and reliability the hot issue for local AI apps.
The local AI darling just got dragged into an internet food fight — and the comments are the main course. A viral post claims Ollama built its fame on llama.cpp — the open-source engine that made AI models run on laptops — while downplaying credit and even skipping license notices for over a year. After a 400‑day silence on a credit issue, a single perfunctory shout-out appeared, and then Ollama pivoted to its own engine built on ggml. The plot twist? Users say the new setup broke things that worked before: structured outputs, vision features, even model downloads. One HN voice sighed: “Just give the llama team proper credit.”
From there, the thread went full soap opera. Power users grumbled that Ollama’s “Docker for AI” vibe felt like a “dockerfile wannabe” that locks your files in a weird folder you can’t move. Another blasted that downloads were “unusable” due to flaky connections. Meanwhile, loyalists clapped back that Ollama is still stupid-simple — type one command, ‘hotswap’ models, done — and that even llama.cpp now has router features closing the gap. One commenter just shrugged, “i use goose by block,” instantly turning into a meme for people peacing out. In short: half the crowd is shouting credit the creators and stop breaking stuff, the rest just want an easy button that works.
Key Points
- •Ollama initially relied on Georgi Gerganov’s llama.cpp for local LLM inference and popularized a simple model management workflow.
- •The article alleges Ollama omitted required MIT license notices and failed to prominently credit llama.cpp for over a year.
- •Community GitHub issues (#3185, #3697) and PR #3700 led to a brief acknowledgment of llama.cpp in Ollama’s README.
- •In mid-2025, Ollama replaced llama.cpp with a custom backend built on ggml, citing stability for enterprise users.
- •The article reports regressions after the switch, including broken structured output, vision model failures, GGML assertion crashes, and incompatibilities like with GPT-OSS 20B.