🔥Beta testing

We are currently in beta testing and would love your feedback before the final release. We would appreciate any suggestion that will help us improve the solution before a wider release.

👉 Share your feedback here. Thanks! It matters a lot.

About Galaxia Graph RAG

Galaxia Graph RAG is a one-stop solution to build graph retrieval model for RAG applications. It is based on a programmable graph (Galaxia) with built-in NLP and a multidomain knowledge corpus.

Core capabilities

• CPU / RAM (in-memory) processing

• Automated knowledge augmentation

• Automated graph construction

• Built-in retrieval

• Modality: text

• File formats: csv, txt

• Context size: 20 million characters (approx. 7000 pages)

Frequently Asked Questions

On what machines is the data processed?

We are currently testing the solution on VM 30GB, 4 vCPU.

How big is the context (text size) I can process at once without chunking?

Galaxia Graph RAG can process up to 20 million characters (~7,000 pages) in a single retrieval.

• This equals approximately 20MB (CSV or TXT).

• Perfect for large-scale graph retrieval without chunking.

Model	Context window (characters)	Approx. page count
Mistral Large 2, Meta Llama 3.2, GPT-4o	512k (~128k tokens)	~200 pages
Google Gemini 1.5	8M (~2M tokens)	~3,000 pages
Galaxia*	20M characters	~7,000 pages

*Please note that Galaxia is not a languge model.

How can I analyze larger text sizes?

At the moment you can do this by processing separate files. After performing the NLP analysis, select all files for which you want to build a graph and choose the 'Build RAG' option. We suggest that the total size of files does not exceed 200 MB for now. Processing larger data sizes will be possible for subsequent iterations of the solution.

Do I need additional embeddings?

No, you don't need additional embeddings. Galaxia has built-in automated knowledge augmentation. It extends the analyzed text with synonyms, taxonomic and ontological relations.

What languages ​​do you support?

Galaxia works for over 20 languages, but for the first versions of Graph RAG, we only allow text processing in English.

My data is in different formats (.pdf, .html etc.)

To find out more, see our Docs processing guides.

Are you compatible with AI agents?

Yes, you can use our solution to build an agentic RAG.

Example combination: LlamaIndex (agents) + Smabbler (Graph RAG / retrieval via API) + aimlapi.com (multiple LLMs via API).

Which functionalities are currently being optimized?

We are currently working on optimization: - retrieval latency, - confidence score, - increasing the size of the analyzed text.