Core Concepts

Understand three core objects and you know Conda's skeleton: environment, channel, package.

Environment

An environment is an isolated workspace with its own Python interpreter and site-packages. Switching environments switches the whole dependency set — the root of how Conda resolves "version conflicts".

• Location: default ~/miniconda3/envs/<name>.
• Analogy: like "a separate Python room" — NumPy 1.x in room A, NumPy 2.x in room B, no interference.

Channel

A channel is the download source for Conda packages, searched from highest to lowest priority. Common channels:

Channel	Notes
defaults	Anaconda-maintained packages (default)
conda-forge	Community-driven, most complete and fastest-updating (recommended)
pytorch	Official PyTorch channel

Enable a channel:

Linux / macOS

conda config --add channels conda-forge
conda config --set channel_priority strict

💡 channel_priority strict makes Conda prefer resolving all deps from one channel, greatly reducing conflicts from mixing channels.

Package

A package is Conda's smallest managed unit, from two sources:

• Conda package: binary built by Conda, may include non-Python deps (e.g. CUDA).
• PyPI package: pure-Python installed via pip, also usable inside Conda envs.

⚠️ Inside a Conda env, prefer conda install; fall back to pip install only when a package is absent. Do not mix pip and conda to uninstall the same package — it breaks the env.

How they relate

   Channel (defaults / conda-forge)
            │ conda install
            ▼
   Package ──create / install──▶ Environment
                                     │ activate
                                     ▼
                            isolated Python interpreter + deps

Install steps: Install Anaconda / Miniconda; env & package commands: Commands.