LLMs1 min read
Local LLMs on a DevOps Workstation
Practical notes on Ollama, quantization, and when local models beat cloud APIs for infrastructure work.
ollamalocal-llmprivacy
Running models locally changes the economics of iteration. You can try prompts against sensitive logs without sending them to a third party.
What works well locally
- Drafting kubectl / Terraform snippets
- Summarizing CI logs
- Rewriting runbook sections
- Exploring MCP tool schemas
What still needs frontier models
- Long multi-repo reasoning
- High-stakes security review
- Complex multi-agent orchestration
ollama pull llama3.2
ollama run llama3.2Start local, escalate to cloud when quality or context demands it.