Just a heads up that if you want to host full local models, you may want to try LM Studio or vMLX with models straight from hugging-face. Ollama is popular but they privilege their proprietary cloud MCP services and won’t even let you download web-enabled models unless you agree to use their API, which defeats your on-prem threat model.
Just a heads up that if you want to host full local models, you may want to try LM Studio or vMLX with models straight from hugging-face. Ollama is popular but they privilege their proprietary cloud MCP services and won’t even let you download web-enabled models unless you agree to use their API, which defeats your on-prem threat model.
Thanks for letting me know, I am still several months away from getting any hardware for a local LLM.