Documentation
Parse, chunk, embed and retrieve documents into a vector store.
Atrium parses PDF, Office documents, HTML, Markdown, EPUB and scanned pages into structure-aware chunks, embeds them with any OpenAI-compatible endpoint, stores them in SQLite, Qdrant or pgvector, and serves retrieval over HTTP.
Two things make it different from the heavyweight stacks:
- Parsing needs no model. The parser and the chunker are pure Python and heuristics. No GPU, no API key, no model checkpoint between your document and the text you get back.
- Chunks keep their provenance. Every chunk knows the page it came from, the heading path it sits under, the element type it is, and the chunking strategy that produced it.
pip install atrium-ragWhat is not here
This is the first public release. There is no distributed job queue, the sparse BM25 ranker is not reachable from any endpoint, and reranking is opt-in and off by default. The limitations page says all of it, including the parts that are uncomfortable.
Where to go next
Quickstart
From an empty directory to a searchable corpus.
HTTP API
Every route, what it takes and what it returns.
The command line
Ten commands, and a setup wizard that asks five questions.
Parsers by format
What each parser produces, and where it gives up.
Chunking strategies
Five ways to cut a document, and what each one costs.
Limitations
What is weak, what is missing, and what the measurements actually say.
Atrium Cloud
Don't want to run it yourself? Atrium Cloud is coming.
It is not here yet — the list is what there is until it is. The library is free, MIT licensed, and yours either way.