Can Denario run on Windows?

[jbdatascience]

Can Denario run on Windows?

I get this error when I try running on Windows:

denario installation blocked: healpy (required by denario → cmbagent) fails to build on Windows due to missing C libraries (cfitsio, healpix)

I am running on Python 3.13.9.

[mikhakuv]

I have the same question, as I tried installing the Denario app on Windows, but ran into version conflicts between Python and one of its dependencies:

• With Python 3.12 'pip install denario' starts correctly, but fails to build wheels for healpy.

• With Python 3.11, healpy can be installed successfully, but 'pip install denario' fails because Python requirement (≥3.12) is not satisfied.

[jbdatascience]

I succeeded to run Denario in a free COLAB notebook! It had no problem installing healpy, because COLAB runs in a Linux environment.

At the end, while generating the paper PDF, it runs into an error, probably caused by some LaTeX problem(s)? I have to work on that.

Also I wonder if one always have to provide a dataset as input. The harmonic oscillator example does not explicitly needs an input dataset. In the description of that example Denario is asked to generate a dataset itself!

Experimenting a lot with Denario can get costly, so watch out for that! I myself are only using OpenAI LLM Models.

I am curious about your results!

[jgooz]

I successfully installed and ran DOCKER, although I encountered some problems during use because I am still learning how to use it.

[jbdatascience]

I successfully installed and ran DOCKER, although I encountered some problems during use because I am still learning how to use it.

I will try DOCKER too. At the moment I can run DENARIO succesfully inside a free COLAB notebook. (This is a good thing: there it can run in an ISOLATED environment so DENARIO can execute Python code it has generated in a safe manner ! Something that also is accomplished by running DENARIO in DOCKER of course).

Up till now I am only using OpenAI Models in DENARIO, which works fine, but no citations are generated because that requires access to the Perplexity API which I will try (I hope there is a free variant of that).

Also I succesfully tried literature research for detecting the novelty of a DENARIO generated idea, although the FutureHouse does not work for me yet (probably also needs API access ?).

In the mean time I am thinking which input problems I will give to DENARIO (and watching the costs of running DENARIO …)

[spark706]

@jbdatascience

Identifying the Root Cause: The healpy Dependency Chain

After troubleshooting the Windows installation, I identified that the primary failure point is the healpy package. While healpy is a powerful tool for cosmology, it does not currently support native Windows environments.

The issue is compounded because cobaya imports healpy. Therefore, even if you don't intend to use the cosmology features, the mere presence of cobaya in the dependency tree triggers a build failure on Windows.

How I Resolved It:
To get the system running natively on Windows (verified as of 2026-01-11), I performed the following:

Cloned and Modified cmbagent: I created a local clone of the cmbagent repository and modified the source code to remove the cobaya and healpy imports. This "decouples" the agent from the Linux-only scientific stack.

Redirected Dependencies: In the main denario pyproject.toml, I pointed the cmbagent dependency to my modified local/GitHub version instead of the PyPI release.

Hatchling Configuration: I added allow-direct-references = true under [tool.hatch.metadata] to permit these Git/local references.

This allows the multi-agent system and the DenarioApp GUI to function perfectly for other scientific domains (like DERMS research) that do not require the cosmology-specific modules.

Native Windows Setup (Without Docker)

For users wanting to run Denario and the GUI on Windows, you can avoid Docker by using the uv package manager. Here is the workflow we used to get it running:

1. Project Configuration
To allow the project to use custom GitHub repositories or local clones, you must first modify the pyproject.toml in the main Denario directory to authorize direct references:

[tool.hatch.metadata]
allow-direct-references = true

2. Customizing Dependencies
If you need to make changes to the app or agents (as I did with cmbagent and DenarioApp), you can clone those repositories locally and point your main project to them.

In your pyproject.toml, update the dependencies to point to the specific Git URLs. This is also the stage where you should remove cobaya and healpy, as they are not currently Windows ready and will cause installation failures.

3. Syncing with uv
Once your pyproject.toml reflects the cloned/customized repositories, run:

# Clear cache to ensure a fresh build
uv cache clean

# Sync the environment
uv sync

4. Running the App
After the sync is complete, you can launch the GUI natively:

denario run

Summary of the Fix
The key was bypassing the standard PyPI versions of cmbagent and DenarioApp by using uv sync with GitHub direct references. This allows the core multi-agent system to function on Windows, though cosmology features requiring healpy still require WSL2 or Linux for full support.

Edit: Forgot to mention that after i make sure denario works for my case with the changes described above, I will make a pull request in the appropriate repos with these changes.