The First Version Was Just a CLI - and That Was the Point

The first working version of Archimedes was not a polished web app.
It was a CLI tool.

And honestly, that was the correct move.

When you are trying to prove a workflow, the CLI is brutally useful.
It strips away all the distractions.
There is no fancy UI pretending to solve the problem for you.
There is just input, output, and the question:
does this actually help?

My early CLI version focused on the core loop:

1. enter research keywords
2. ask a research question
3. fetch papers
4. run analysis
5. print a structured answer

That was enough to learn a lot.

I found out quickly that the magic was not in making the tool look impressive.
The magic was in whether the pipeline felt useful.
If the CLI could turn a messy research prompt into something structured and evidence-based,
then I had something worth keeping.

The CLI also forced discipline.
It made me think about the minimum useful workflow.
What is the smallest version of this idea that still feels like a real product?

That led to the current shape of Archimedes:

• search arXiv and OpenAlex
• deduplicate and normalize papers
• analyze each paper with an LLM
• aggregate findings
• synthesize a direct answer
• export the result as a PDF

A CLI is not sexy.
It is not marketable in the same way a shiny website is.
But it is perfect for proving that the thing actually works.

Once the CLI felt useful for me, I knew the next step was not "make it prettier".
The next step was "make it accessible."

That is where the web version came from.