Technology

How to Install Miniconda Safely on Windows (Terminal Only), and When to Use Pixi or uv Instead

A safe, terminal-only Miniconda setup for Windows: no PATH mess, no surprise licence fees, conda-forge instead of Anaconda's paid channel. Plus an honest comparison with pixi and uv, their limitations, and which one fits your project.

If you're getting into data science, machine learning or any kind of scientific Python, someone will eventually tell you to "just install conda". That's usually good advice. But there are two traps nobody mentions: the installer can quietly mess up your other Python setups, and the default package source comes with licence terms that can cost companies money.

This guide installs Miniconda the safe way, entirely from the Terminal, with no clicking through installers. After that, I'll explain the two modern alternatives, pixi and uv: what they're good at, where they fall short, and which one I'd pick for different kinds of projects.

You don't need to be an expert. Every command is explained, and you can copy and paste them straight into the Terminal.


First, What Is Conda and Why Do People Use It?

Python packages are add-ons that other people wrote, like pandas for spreadsheets or numpy for maths. Most of them install fine with Python's normal tools. But some packages depend on things that aren't Python at all: compiled maths libraries, GPU drivers, mapping software, even the R language. Getting those working by hand on Windows is painful.

Conda solves that. It's a package manager that can install both Python packages and the non-Python pieces they need, and it keeps each project in its own separate "environment" so they never clash.

Miniconda is the small, minimal installer for conda, made by the company Anaconda. It gives you conda and Python, and nothing else, so you only install what you actually need. (The full "Anaconda Distribution" comes with hundreds of packages preinstalled and takes up several gigabytes. Most people don't need it.)


The Licence Catch (Read This Before Installing)

Here's the part that trips people up. The Miniconda software is free, but by default it downloads packages from Anaconda's own package channel, and that channel has terms of service. According to Anaconda's own explanation in "Is conda free?", it's free for individuals and for organisations with fewer than 200 people. Larger organisations generally need a paid licence, with some exceptions such as certain academic uses.

Since 2025, newer versions of conda also ask you to formally accept these terms before using Anaconda's channel. That's what the `conda tos` command is for.

The good news is that there's a completely free alternative source called conda-forge. It's run by the community, has a bigger and often more up-to-date package collection, and is free for everyone, including big companies. So the "safe" setup below switches conda over to conda-forge.

An even simpler option is Miniforge. It's the community's free version of Miniconda, set up to use conda-forge from the start, with no Anaconda channel and no terms to accept. If you're installing this on a work laptop, I'd honestly just use Miniforge. I've included the command for both below.


What Makes This Setup "Safe"

The default installation choices are where most conda headaches come from. This guide makes four safe choices:

  1. Installed for your user only. No administrator rights needed, and it can't affect other people's accounts.
  2. Not added to your system PATH. Adding conda to PATH can hijack the python command for everything else on your computer, including other Python installs and tools like uv. We'll connect it to the Terminal properly instead.
  3. Not registered as the system's default Python. Other apps won't accidentally start using conda's Python.
  4. Installed with winget, Windows' built-in installer. It checks each download's digital fingerprint before running it, so you know the file hasn't been tampered with.

Step 1: Open the Terminal

Right-click the Start button and choose Terminal. Everything below runs in this window. Copy a command, paste it with Ctrl + V, and press Enter.


Step 2: Install Miniconda (or Miniforge)

For Miniconda, run:

winget install --id Anaconda.Miniconda3 -e --override "/InstallationType=JustMe /AddToPath=0 /RegisterPython=0 /S /D=$env:USERPROFILE\miniconda3"

Or, for Miniforge (the fully free, conda-forge-only version), run this instead:

winget install --id CondaForge.Miniforge3 -e --override "/InstallationType=JustMe /AddToPath=0 /RegisterPython=0 /S /D=$env:USERPROFILE\miniforge3"

The part in quotes tells the installer exactly how to set itself up, instead of using its defaults. In plain English:

  • /InstallationType=JustMe: install for my account only.
  • /AddToPath=0: don't touch my system PATH.
  • /RegisterPython=0: don't make this my computer's default Python.
  • /S: install silently, without opening any windows.
  • /D=...: install it into a folder in my user profile. This one has to come last.

These are the official installer options documented in the conda installation guide for Windows. The install takes a minute or two. When it finishes, close the Terminal and open a new one.


Step 3: Connect Conda to the Terminal

Because we deliberately didn't add conda to PATH, we need to tell PowerShell about it once. First, allow PowerShell to run the small setup script conda adds to your Terminal profile. This setting only applies to your user account:

Set-ExecutionPolicy -Scope CurrentUser RemoteSigned

If it asks you to confirm, type Y and press Enter. Then run conda's setup:

& "$env:USERPROFILE\miniconda3\Scripts\conda.exe" init powershell

(If you installed Miniforge, replace miniconda3 with miniforge3 in that command.)

Close the Terminal and open a new one. Now check that it works:

conda --version

You should see a version number like conda 26.x.x.


Step 4: Stop Conda Taking Over Every Terminal

By default, conda switches on its main environment, called base, every time you open a Terminal. You'll see (base) at the start of each line. That sounds harmless, but it means every Terminal window is silently using conda's Python, which is confusing and makes it easy to install things in the wrong place.

Turn it off:

conda config --set auto_activate false

(On older conda versions, the setting is called auto_activate_base. If the command above gives an error, run conda config --set auto_activate_base false instead.)

From now on, conda only switches on when you ask it to. It's a small change that saves a lot of confusion.


Step 5: Switch to the Free conda-forge Channel

Skip this step if you installed Miniforge. It already uses conda-forge.

For Miniconda, run these three commands:

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

The first adds conda-forge as a package source. The second tells conda to strictly prefer it, which avoids mixing packages from different sources (a common cause of strange errors). The third removes Anaconda's own channel. Removing that channel is the method Anaconda's documentation itself recommends if you don't want to use it. All the settings are explained in the conda configuration docs.

One honest note: the packages that came with the Miniconda installer itself were originally downloaded from Anaconda's channel. If you're an individual or at a small organisation, that's within the free terms. If you're at a large company, this is exactly why Miniforge is the cleaner choice.


Step 6: Create Your First Environment

Never install your project packages into base. Instead, create a separate environment for each project. Here's one for data analysis:

conda create -n data python=3.14 pandas numpy jupyterlab

Conda shows you what it's going to install and asks you to confirm. Type y and press Enter. Then switch the environment on and check it:

conda activate data
python --version

When you're done working, switch it off with conda deactivate.

To share your environment with someone else, or rebuild it later, save its recipe:

conda env export --from-history > environment.yml

The --from-history part saves only the packages you actually asked for, not hundreds of sub-packages, so the file also works on other computers.


The Alternatives: pixi and uv

Conda isn't the only option any more. Two newer tools have become very popular, and depending on your project, one of them might suit you better.

pixi: conda's packages, done the modern way

pixi is a free, open-source tool made by a company called prefix.dev. It installs packages from conda-forge, so you get all of conda's non-Python superpowers, but it works per project instead of with named environments. That means no Anaconda channel and no terms to accept.

What I like most about pixi is that it does two things conda makes you do manually. First, it automatically writes a lock file that records the exact version of every package, so your project installs identically on any computer, months later. Second, it lets you define tasks, like pixi run test or pixi run start, so everyone on a team runs things the same way. It's also very fast. It can install regular PyPI packages alongside conda ones, too.

Install it with winget:

winget install prefix-dev.pixi

Then, after reopening the Terminal, starting a project looks like this:

pixi init my-analysis
cd my-analysis
pixi add python pandas numpy
pixi run python --version

The pixi installation docs cover other install methods.

Its limitations: pixi is younger than conda, so there are fewer tutorials, and most online answers still assume conda commands. Its project-first approach is different if you're used to conda activate with named environments that you reuse across projects (pixi has pixi global for installing standalone tools, but it's not the same thing). And some university courses and company setups simply expect conda, so you may not get a choice.

uv: the fastest option, for "normal" Python

uv is a free tool from Astral, and for most everyday Python work it's the one I'd recommend first. It installs Python itself, creates a private environment for each project, writes a lock file, and it's incredibly fast. I've written a full beginner's guide to setting up Python with uv on Windows.

Its big limitation: uv only installs packages from PyPI, Python's standard package index. It can't install the non-Python pieces that conda and pixi handle, like specific GPU toolkits, system libraries or R. In practice, that matters less than it used to. Most popular packages, including pandas, NumPy and even PyTorch, publish ready-made versions that bundle what they need. But for some scientific and geospatial libraries, you'll hit a wall that conda or pixi would have handled for you.


Which One Should You Use?

Here's how I'd decide, based on the kind of project you're working on.

For web apps, APIs, automation scripts and learning Python, use uv. These projects almost never need non-Python libraries, and uv is the fastest and simplest of the three. This covers Django, Flask and FastAPI backends, bots and command-line tools.

For data science and research with heavier dependencies, use pixi. That includes geospatial work (GDAL and other mapping libraries), bioinformatics, projects mixing Python and R, and setups that need a specific GPU toolkit. It's especially good when a team needs everyone's environment to be identical. You get conda-forge's huge package collection with modern conveniences, and no licence questions.

Use Miniconda or Miniforge when your course, team or computing cluster already uses conda. If your instructions say conda install, it's usually easiest to follow them, especially on shared university servers where conda is often the standard. And between the two, pick Miniforge unless you specifically need Anaconda's channel. It's the same conda, without the licence fine print.

You can also mix them. Using uv for web projects and pixi for the occasional heavy data project on the same laptop works perfectly well. Because we set conda up without touching PATH, none of them get in each other's way.


Common Problems

conda is not recognized You haven't reopened the Terminal since running conda init. Close every Terminal window and open a new one. If it still fails, check that the folder name in Step 3 matches where you installed it (miniconda3 or miniforge3).

"running scripts is disabled on this system" The execution-policy command in Step 3 didn't apply. Run Set-ExecutionPolicy -Scope CurrentUser RemoteSigned again, then open a new Terminal.

Conda asks you to accept Terms of Service You're still using Anaconda's channel. If you're eligible to use it for free and want to keep it, accept with conda tos accept. Otherwise, follow Step 5 to switch to conda-forge, or use Miniforge.

Installs are painfully slow or end in "conflicts" You're probably mixing channels. Make sure Step 5 is done (strict priority, defaults removed), and create a fresh environment rather than piling more packages into an old one.


How to Uninstall It Cleanly

If you decide conda isn't for you, uninstall it from the Terminal:

winget uninstall --id Anaconda.Miniconda3

(Or CondaForge.Miniforge3 for Miniforge.) Then remove the setup lines conda added to your Terminal profile, so PowerShell doesn't look for it on startup. Open the profile file with notepad $PROFILE, delete the block marked #region conda initialize, and save.

That's it: a clean, safe conda setup, and a clear idea of when to reach for pixi or uv instead.