Retrieval-Augmented Generation (RAG) has become one of the most useful machine learning technologies in my daily work as a civil engineer, even as LLM context windows continue to grow. That’s why I created VERA (Vector Embedded Retrieval Archive), a portable RAG format with a CLI that both humans and agents can use.
Github: https://github.com/dkylewillis/vera
Desktop app: Available from the GitHub Releases page.
It’s as simple as:
pip install "vera-cli>=0.3.1"
vera convert manual.pdf manual.vera
vera search manual.vera "when is stormwater detention required?" --json
Each PDF becomes its own self-contained .vera archive: a SQLite 3 database containing chunks, embeddings, a keyword index, metadata, attachments, and the original PDF. You can search one archive, multiple archives, or index an entire library to search thousands of documents like a vector database while keeping each document portable.
At work, I’ve converted and indexed more than 1,200 contracts. My Grok Bot has a VERA skill that searches them for relevant context and can draft new contracts based on my instructions while following our company’s historical style.
I also built an application around VERA for more hands-on research. It includes visual grounding and an LLM interface constrained to retrieved .vera context, with visual citations that take you back to the actual source language.
Researching codes, ordinances, contracts, and technical manuals is part of my daily work. RAG lets me do that due diligence much faster, while VERA gives me a way to verify the LLM’s answers against the original source.
If you have a use for something like this, I’d be interested to hear how you use it.
