Skip to content

Overview

ModelDock is a lightweight, Python-first model management layer for local LLM runtimes — the "package manager for local AI models."

What ModelDock Does

It manages discovery, download, caching, installation verification, and loading through pluggable runtime adapters. It does not run inference itself; it orchestrates runtimes (starting with Ollama).

Core Concept: The load() Flow

The flagship operation:

Model requested | Already installed? ──yes──> return client | no Download automatically (with progress) | Verify installation (checksum + runtime check) | Load model -> return ready client

Key Capabilities

Capability Description
Auto-install load() downloads missing models automatically
Searchable catalog Browse models, categories, capabilities from Python
Bulk install install_category("coding") pulls related models
Smart caching Never re-download installed models
Runtime adapters Ollama now, LM Studio/llama.cpp/vLLM later
Cross-platform Windows, macOS, Linux via platformdirs
Zero-config Dynamic catalog from ollama.com with offline fallback

Architecture Layers

ModelDock follows Clean Architecture with SOLID principles:

Layer Location Purpose
Interface modeldock/__init__.py + modeldock/cli SDK and CLI
Application modeldock/core/ Services, orchestrator, manager
Domain modeldock/domain/ Pure entities, no I/O
Ports modeldock/ports/ Protocol interfaces
Adapters modeldock/adapters/ Runtimes, registry, downloaders
Common modeldock/common/ Config, logging, errors

Dependencies point inward: cli → core → ports ← adapters.

Next Steps