Orientation: a coding model on your own laptop

1. The promise

You'll know what this chapter delivers, what "local" really means here, what hardware you need, and where this fits in the larger course. By the end of the chapter you'll refactor a real module with a model running on your own laptop.

2. The mental model

Think of this chapter as installing a workshop bench in your own garage. A hosted model is renting time at the workshop downtown. Convenient, well-equipped, but somebody else's space, somebody else's bill, somebody else watches what you make. A local model is your own bench. Cheaper to run forever, private, and good enough for most jobs. The fancy workshop downtown is still there for the jobs your bench can't handle.

(DIAGRAM: local vs hosted inference paths — flowchart — top path: prompt → Continue → Ollama → model on disk → answer, all inside one labelled "your machine" box. Bottom path: prompt → editor → network → cloud provider → model on their servers → network → answer)

A few words to anchor before we go further:

• Inference — the act of running a trained model on a prompt to produce an answer. Not training, not fine-tuning. Using.
• Weights — the numbers that define a trained model. A file on disk, often several gigabytes.
• Runtime — the program that loads the weights and runs inference. Ollama is one runtime; there are others.

3. You do it, with me

Open VSCode. You should have it from Ch 0.

(SCREENSHOT: the Click to Career chapter index showing Ch 22 — full window — Ch 22 highlighted)

Open Activity Monitor on macOS or Task Manager on Windows. Note your current memory use with no AI tooling running. That number is your baseline.

(SCREENSHOT: Activity Monitor (macOS) or Task Manager (Windows) before any AI tooling is installed — full window — Memory tab visible — showing baseline RAM usage)

Open a browser tab and visit ollama.com. Don't download yet. The install walkthrough is in 22.3.

(SCREENSHOT: the Ollama website homepage at ollama.com — full window — no highlights)

The four tools the chapter installs:

• Ollama — the runtime that loads models and answers prompts on your machine.
• Continue.dev — the VSCode extension that turns Ollama into chat, autocomplete, and inline edit inside the editor.
• Cline — the alternative VSCode extension for agent-style work that can call tools and run shell commands.
• One real coding model chosen at install time.

The hardware reality, briefly. 16 GB of unified memory (Apple Silicon) or 16 GB RAM with a recent discrete GPU is the comfort floor. Less than that works for autocomplete-class models, not for chat-class models. The detail lands in 22.7.

Course context. This chapter is part of Block 7 (AI tooling). The next chapter, Ch 23, introduces Claude Code — the harness this course's author uses for the bulk of the writing. Local AI first, hosted AI second. The order matters.

One-line policy: from here on, a local model is your default for code work. Hosted models are for the work the local model can't finish.

4. What you should be seeing

• VSCode is open.
• A browser tab is open at ollama.com.
• You have a baseline memory number in your head for your machine before any AI tooling runs.
• You can say in one sentence what a local model is and what a hosted model is.

If any of those four are missing, go back and fix it before 22.1.

5. Common stumbles

• You assume "local" means "offline" the moment Ollama is installed. The model download happens over the network. After that, the workflow is offline.
• You expect a 7B local model to feel like Claude. It won't. A 7B-parameter model is good enough for most refactors, not for hard reasoning. Set expectations now.
• You want to skip to Cline or Claude Code without learning Ollama. The runtime is the foundation. Every other tool sits on top.

6. You should know