AI splits into two halves, and the distinction matters. Half of these run the model yourself: LLM inference and fine-tuning tools, machine-learning frameworks. The other half use a model's capability: chat clients, agent frameworks, AI coding assistants, image tools, speech-to-text, AI voice tools, second-brain and knowledge-base apps, MCP connectors.
The first group wants a GPU, VRAM and downloaded model weights, plus environment setup afterwards. Most of the second group needs only an API key, which is a much lower bar. Which path a given need takes depends on whether you are willing to run inference locally.
If you just want AI to do the work, start with AI Chat Clients and AI Coding Assistants. To run models on your own machine, look at LLM Inference and Fine-Tuning, and check VRAM requirements and quantisation options first. To point AI at your own files and data, look at MCP Connectors and Second Brain / Knowledge Base.
