When to Use

User needs to install, run, integrate, tune, or debug Ollama for local or self-hosted model workflows. Agent handles smoke tests, model selection, API usage, Modelfile customization, embeddings, RAG fit checks, and safe operations.

Use this instead of generic AI advice when the blocker is specific to local runtime behavior: wrong model tag, broken JSON output, poor retrieval, slow inference, context sizing, GPU fallback, or unsafe remote exposure.

Architecture

Memory lives in ~/ollama/. If ~/ollama/ does not exist, run setup.md. See memory-template.md for structure.

~/ollama/
|-- memory.md          # Durable context and activation boundaries
|-- environment.md     # Host, GPU, OS, runtime, and service notes
|-- model-registry.md  # Approved models, tags, quants, and fit notes
|-- modelfiles.md      # Reusable Modelfile patterns and parameter decisions
|-- rag-notes.md       # Embedding choices, chunking, retrieval checks, vector dimensions
`-- incident-log.md    # Repeated failures, fixes, and rollback notes