You've installed Ollama at some point. Forgot about it. This guide gets it into a real workflow.
Step 1: install the Workbench and point it at http://localhost:11434. If Ollama isn't running yet, npm start does nothing until the service is listening.
Step 2: pull two models you actually trust. We recommend one capable model and one fast model. Ask the first a hard question. Ask the second to summarise the last hour. Compare the answer quality — you'll find they're both usable for different jobs.
Step 3: attach a real document. A PDF, a spec file, or a contract. Ask your question. Because the whole pipeline lives locally, nobody writes your private documents to anyone's log.
The habit that forms fastest: you stop copy-pasting into cloud chat and start trusting local answers for day-to-day work.