Ollama and Continue: local coding setup in VS Code

This tutorial connects Ollama to the Continue extension in VS Code for a local coding assistant. The speaker's goal is to reserve cloud tools for complex work and use smaller models for routine questions and code exploration, reducing paid token consumption.

The setup covers installing Ollama on a Mac, downloading a model with ollama pull, and configuring Continue manually with Ollama as the provider. The speaker sets Ollama's context window to 128k, then demonstrates Continue's agent, plan and chat modes. An initial request supplies a Swift file as context and asks the model to explain it.

The longer exercise asks an AI agent to locate and fix a broken timer animation in an iOS app. The model finds relevant files and proposes edits, but the changes produce build errors. Repeated attempts with compiler logs do not resolve the problem. The speaker then switches from the model called Gemma 4 to one transcribed as "Queen 3.6," downloaded as 22.3 GB.

During that second attempt, the speaker reports 23.39 GB of memory use by the Ollama server on a Mac with 32 GB RAM, alongside Xcode and a simulator. Those figures describe this session, rather than a universal hardware requirement. The final assessment is that the tested models helped with smaller tasks but did not complete this iOS repair.