Originally published at: Meet editorial-guide-ramalama: An AI Assistant That Checks Your Fedora CommBlog and Magazine Articles Against Editorial Guidelines – Fedora Community Blog
Introduction
Open source communities run on contribution, and contribution runs on documentation, storytelling, and knowledge sharing. At the Fedora Project, that means the Fedora Community Blog and Fedora Magazine, two publications that give contributors a voice and give the community a way to stay informed, inspired, and connected.
This year, as part of my internship with the Fedora CommOps team at Red Hat Ireland, I got to experience that firsthand. Publishing 11 articles, running editorial campaigns, and coordinating content across sprints gave me a deep appreciation for how much thought goes into keeping Fedora’s editorial standards consistent and how much easier that journey could be for new contributors with the right tool in hand.
That’s exactly what my fellow intern Gonothi built during his Outreachy internship with the Fedora Project. In our final sprint together, we decided to share it with the community because tools that make contributions more accessible deserve to be heard. With that context in mind, I’ll hand over to Gonothi to walk you through what he built and why.
What I built and why
My name is Gonothi, and I’m an Outreachy intern within the Fedora Project. The tool I built is called editorial-guide-ramalama, a Retrieval-Augmented Generation (RAG) assistant that reads Fedora’s actual editorial guidelines and published articles, then checks your draft against them. It tells you whether your article meets the standards and exactly what to fix if it doesn’t.
A regular chatbot would guess. RAG grounds every answer in the actual guidelines. When the tool flags a problem, it cites the specific guideline and gives you an actionable fix. Not “this looks off”- but “your article is missing the Read More tag, which is required per the Magazine guidelines.”
RamaLama is the engine behind the whole tool, and it’s a big part of why this project works the way it does. It runs open models locally as OCI containers, the same container tooling Fedora already uses, so there’s no API key, no external service, and no data leaving the machine, which keeps everything private and fully reproducible. It also has RAG built in: RamaLama handles the ingestion, chunking, and retrieval itself (running Docling internally to parse and chunk the guidelines), so I don’t have to wire together a separate vector pipeline. That combination of local inference, OCI-container packaging, and built-in RAG is what lets the tool ship as a single image that anyone can pull from Quay and run.
How it works
The tool supports both Fedora publications, Fedora Magazine and the Fedora Community Blog, each with its own editorial guidelines loaded as the primary source the model checks against. Switch publications in the sidebar, and the guidelines change accordingly.
The stack is built on RamaLama, Docling, Quay, and Streamlit, with small GPT-Generated Unified Format (GGUF) models benchmarked for quality versus size. The pipeline works like this:
Articles are pulled from both publications via the WordPress REST API; no static files are committed, and the corpus is always rebuilt from source and stays reproducible. RamaLama RAG runs Docling internally to parse and chunk the documents into a vector store, packaged as OCI images published to Quay at quay.io/fedora/editorial-guide-ramalama. To run a check, pull the relevant image and use either the Streamlit interface or the terminal; no build step required.
Paste a draft that meets the Magazine’s standards, and the model confirms compliance, citing the relevant guideline. Paste one with known issues, a missing featured image, a missing Read More tag, and it flags each problem with an actionable fix.
One honest note: small local models have limits. editorial-guide-ramalama is great for catching common issues before submission. It is not a replacement for a human editor, and it doesn’t try to be.
Why the open source community should care
What strikes me most about this project is not just what it does technically, but what it represents for contributor onboarding in open source communities. One of the biggest barriers for new Fedora contributors isn’t motivation; it’s knowing the unwritten rules. Editorial guidelines, packaging standards, community norms these exist, but they’re scattered, and learning them through rejection is discouraging.
Tools like editorial-guide-ramalama lower that barrier. They don’t replace human editors or community knowledge; instead, they give new contributors a first pass, a way to self-check before they submit, and a way to learn the guidelines through doing rather than through trial and error. That’s exactly the kind of tooling that makes open source communities more accessible.
What comes next
This mini-project was the proof of concept that validated the RAG approach for Fedora editorial documentation. The same architecture is now feeding into a larger project applying RAG to RPM packaging guidelines, a higher-stakes domain where grounded, local, open-model AI can help contributors get packaging right.
The code is on the Fedora Forge at ai-ml/editorial-guide-ramalama. Pull the image from Quay and try it on your next draft before you submit.
Try it yourself
The tool is live and available now. Pull the image from Quay, paste your draft, and see what it says. If you’re a new contributor, nervous about your first article, this was built for you.
Thank you to the Fedora community, our mentors Justin Wheeler, Dominik Kawka, and Carol Chen, and Outreachy for making this internship possible.