Repository · RAG
Your documents,
answering back.
A document store that doesn't insist on owning your files. Keep a pointer to what already lives in SharePoint or S3, write new material in place, then vectorize any of it into a collection the Reactor can answer from, with the source cited on every claim.
From file to grounded answer
Four steps, no pipeline to build.
The chunk, embed and store work is one action on a document or a whole folder.
Add
Upload it, reference it where it lives, or write it in place.
Chunk & embed
Split into passages and embed each one.
Store
Vectors land in pgvector by default, or ChromaDB.
Ask
A RAG target System answers in the Reactor, with citations.
Three kinds of document
Keep the file where it already lives.
Upload
physicalQuadrazene stores the bytes itself. Use it when there is no other home for the file.
Reference
no second copyStore only a pointer to a file in SharePoint, S3, Azure Blob or GCS. Nothing is copied; content is fetched from the source each time it is previewed or vectorized.
Author
in placeWrite a Markdown or HTML document directly in the app. Useful for playbooks and glossaries that have no other source of truth.
Copy in, on purpose
Decide later that a referenced file should be held locally? Copy it in. It is always a deliberate, explicit action, never automatic.
Folders
A familiar folder tree with breadcrumbs. Vectorize a single document or a whole folder in one action.
Governed like everything else
Day-to-day document management is open to general users; vectorizing is its own capability that an admin grants.
Retrieval you can defend
An answer with its receipts.
Zero-config vector store
Vectors go into pgvector on the deployment's own Postgres by default, with nothing extra to run. Point at ChromaDB when you want a dedicated store.
RAG target Systems
A collection becomes a System the Reactor can route to, so a policy question goes to your policies, not to a model's memory.
Cited by construction
Answers are synthesized from retrieved passages and carry the passages' sources with them, so a reader can open the original.
Composable in Formulas
Retrieval and answer synthesis are atoms, so a Formula or Chain can ground a step in your documents before it acts.
Trust Layer applies
Answer synthesis is an ordinary governed LLM call: the same content filters, model allowlist and classification ceiling apply as anywhere else.
Pin the answer
Pin a good grounded answer to your Home Watchlist, or a Doc Pin widget to the document itself.
Point it at your policies.
Bring a folder of documents and we'll show the same questions answered with and without grounding.