Skip to content

Vision

ModelDock is the package manager for local AI models.

The Problem

Local LLMs are powerful, but managing them is painful. Developers must manually:

  • Visit the Ollama website to find models
  • Understand variants, sizes, tags
  • Download each model individually
  • Repeat for every new model
  • Manage updates and storage

This interrupts development and creates unnecessary friction.

The Solution

ModelDock provides a lightweight, Python-first interface for discovering, downloading, managing, caching, and loading local LLMs.

Instead of:

ollama pull llama3
ollama pull qwen3
ollama pull deepseek-r1

Just:

import modeldock as md
client = md.load("llama3")

Design Goals

  • Lightweight — no bundled models, minimal dependencies
  • Python-first — clean API, type hints everywhere
  • Beginner-friendly — zero config, smart defaults
  • Smart caching — never re-download installed models
  • Extensible — new runtimes are drop-in adapters
  • Cross-platform — Windows, macOS, Linux
  • Developer-focused — fast, composable, predictable

What ModelDock Is NOT

  • It does not run inference
  • It does not bundle models
  • It does not replace Ollama — it orchestrates it

Future Vision

  • Automatic model downloads
  • Searchable model registry from Python
  • Bulk installation by category
  • Model recommendations for beginners
  • Download progress tracking
  • Version management and updates
  • Multiple runtime support (Ollama, LM Studio, llama.cpp, Jan AI, GPT4All, vLLM)
  • Smart aliases (load("llama3"))

Author

Created and maintained by Himanshu kumar (OpenAgentHQ).