We map knowledge like the stars
Startrag began with a simple frustration: vector search retrieves neighbors, but understanding lives in the connections between them.
Traditional RAG treats your knowledge base as a bag of disconnected chunks. When a question spans several documents, flat vector search returns the nearest neighbors — and misses the relationships that actually hold the answer.
Startrag organizes those chunks into a graph: each piece of knowledge is a star, related stars form clusters, and clusters connect into constellations. Retrieval traverses that structure, so the model sees coherent context instead of fragments.
What we believe
Grounded by default
Every answer should trace back to a source. We optimize for truth, not fluency.
Structure beats scale
A well-mapped graph retrieves better than a bigger pile of vectors.
Developer-first
A small, predictable API you can reason about — no black boxes.