I got tired of updating docs by hand, so I built a one-file Python script

Keeping project docs up to date is one of those tasks that always sounds reasonable until you actually have to do it.

I kept ending up in the same loop: change the code, forget to update the docs, remember later, then spend way too much time rebuilding the project context for myself or for an AI chat.

At some point, I got tired of it and built a tiny open-source script for my own workflow.

The idea was simple:

one Python file, zero dependencies, one command, two output files

It generates:

• CodebaseDump.md — a full markdown snapshot of the codebase that is easy to paste into AI chats
• ReadmeDev.md — a technical developer doc generated from the repository

Why this design?

Because I did not want to build a huge agent system just to solve a boring problem.

I wanted something that:

• works with plain Python
• is easy to drop into any project
• does not need extra setup
• stays compact and hackable
• can still work on free or rate-limited models

So I kept the AI layer simple too.

Instead of asking one model call to do everything at once, the script uses two passes:

• a focused research pass that extracts grounded project context from the codebase dump
• a final aggregator pass that turns that into a usable technical doc

That makes the whole thing faster, cheaper, and more realistic to use with smaller models.

Honestly, that part mattered a lot to me.

The slowest part of tools like this usually is not Python.

It is waiting on model responses.

So I tried to avoid unnecessary AI overhead and keep the workflow practical for people using free keys, small models, side projects, internal tools, or research prototypes.

It is not trying to be the ultimate autonomous documentation agent.

It is just a useful little script that lets me run one command and get:

• a repo dump for AI chats
• a readable technical doc for the project

Good enough to save me from doing this manually over and over again.

Could this become a much bigger agent system later? Sure.

But I actually think there is value in starting with the boring version first:

one file, one command, no dependency drama

That is the version people might genuinely use.

The project is open source here: Bakunin-dev/SumAI

If this sounds useful for your workflow, feel free to explore it, fork it, or make it better.

---

Yes — this post was created by AI.

Praise the Omnissiah. ⚙️