Skip to content

Cache Management

ModelDock tracks what's installed and cached so you never re-download.

Check Cache Status

import modeldock as md

entries = md.cache.status()

Get Cache Path

path = md.cache.path()

Clean the Cache

Safe by default — only removes corrupt/partial entries:

md.cache.clean()

Force wipe everything:

md.cache.clean(force=True)

How Caching Works

ModelDock maintains two cache concepts:

Installed-Model Cache

Tracks models already pulled into the runtime (Ollama's store). Verified via runtime.list_installed() plus a local manifest.

Download Artifact Cache

For runtimes that download raw files (llama.cpp GGUF, GPT4All), a content-addressed store (cache/<sha256[:16]>/model.gguf) makes re-installs instant offline.

Smart Caching Logic

  • CacheService.is_fresh(ref) compares requested tag/spec against manifest + runtime state
  • Content hashing (SHA-256) makes the cache path-independent and self-validating
  • cache.clean() removes partial downloads and orphaned artifacts

Offline Mode

If auto_install is off or network is unavailable, load() fails fast with a clear "model not installed and offline" message rather than hanging.

Next Steps