Hi and welcome !
Regarding your ask for uncensored models, here are a couple of options for your hardware:
Hermes-3-Llama-3.1-70B-Uncensored: If you are looking for a large model that can run on 128GB. You can find it here: Guilherme34/Hermes-3-Llama-3.1-70B-Uncensored
Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive: Personally, I have used this one, but then stayed with the ‘normal’ (not uncensored) version. Half of the size and number of parameters as the Hermes 70B, but works great for me as the brain for Codex agents across multiple clients. On a system with 128GB of RAM (Asus DX G10 in my lab), it leaves a quite large context window of 256K tokens. Get it from: HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive
For your security and telemetry concerns, I would go with the usual layered security approach: Enjoy the new local lab setup!