Hybrid workflow: local for private and repetitive, cloud for hard

What they built

Korbin Brown self-hosts everything he can and hoped Ollama models would replace ChatGPT and Claude. After months of trying, he concluded they can't on heavy tasks, so he now delegates the work local models are actually good at.

Setup

• Ollama with several models
• A local Llama model for summarization
• A lightweight Qwen in VS Code via the Continue extension
• A local Qwen2.5-Coder for code (compared with Claude)

How it works

1. Private work (sensitive documents, contractually confidential code) stays local
2. A Python script feeds saved articles to a local Llama overnight for summaries
3. Continue provides low-latency autocomplete from a small local Qwen

Gotchas

• Multi-step tasks and nuanced writing fall apart fast
• Context retention in long conversations is weak
• Qwen2.5-Coder needed arguing with compared to Claude

“private and repetitive tasks stay local, and complex and important jobs go to the cloud.”— Korbin Brown, XDA

Do it yourself

1. Write a nightly script that sends saved articles to a local Llama model for summaries
2. Install Continue in VS Code and set autocomplete to a small local Qwen coder
3. Keep sensitive documents and code on the local model by default
4. Use cloud models only when a wrong answer has real consequences