Another neat trick, use adversarial local models for security auditing, like jimscard/blackhat-hacker, jimscard/whiterabbit-neo, ALIENTELLIGENCE/cybersecuritythreatanalysisv2 or similar. Just run them in your local GPU on ollama and use it for things like code review and exploit discovery.

The Grok model is powerful. It nailed several things that others seemed to stumble on. If you have cursor installed, just switch to the Grok one and ask it to do a code review and define next steps and gaps and see what you get. It is fast and free, which saves you credits on the other LLMs.

I am also looking at setting up two local AI servers here at the house to offer me unlimited access to okay-ish models to fill the gaps. I have one with a 4090 and 3080 and another I am building with 4x 3080s. Just need time to get it all together and I am chasing fiat for bills rn.