Category: Local AI & LLMs

Running a language model on your own machine changes what you can do with it. There is no per-token bill, no rate limit, and nothing leaves the computer — which is what makes local models usable for client documents, internal notes and anything else you would not paste into a hosted chat window. This section covers that shift end to end: which runtime to pick, how to install it, what hardware actually delivers, and how to put the model behind an API so your own tools can call it. The benchmarks here are measured, not quoted. Throughput, time-to-first-token and VRAM figures come from a single AMD Radeon PRO W7900 with 48 GB, running the models described in each article, so the numbers can be compared against each other. Where a result was disappointing, it is written down as it was measured. Start with the offline LLM guide if you are new to this, then move to the runtime that fits your setup — llama.cpp for control, Ollama or LM Studio for convenience. Once a model runs, the API articles show how to point editors, agents and scripts at it, which is the point at which a local model stops being an experiment and starts replacing a paid service.