EXPLAINERAI & Automation
Ollama Memory Has Three Budgets, Not One
A model download is not a memory budget. Understand loaded models, context length, and parallel requests before sizing local AI for a shared home server.
Read article
Field Notes / Category
Practical agents, MCP tools, routines, monitoring, and safe automation for self-hosted systems.
EXPLAINERA model download is not a memory budget. Understand loaded models, context length, and parallel requests before sizing local AI for a shared home server.
TUTORIALRun Ollama without publishing port 11434, prove an approved Docker client works, and confirm an unrelated container cannot resolve the service.
SECURITYMove one Docker Compose password into a file-backed secret, limit service access, test allowed and denied paths, and keep a safe rollback.
SECURITYDeploy a pinned Ollama container with host-only and container-only access, then prove that LAN and internet paths stay closed.