Five workloads. One standalone database.
Each application is an independent client of KynetraDB. The interfaces are live; scale numbers remain targets until a recorded benchmark proves them.
Kynetra NotesOffline sync
Local cache, queued mutations, deterministic conflict handling
Seven-day offline target; zero acknowledged edits lost 02Kynetra PagesBlock documents
Editable blocks, workspace navigation, entity persistence
Realtime co-edit protocol remains a release gate 03Kynetra GridOperational records
Editable typed rows, formulas, filters, entity persistence
One-million-row indexed view remains a benchmark target 04Kynetra ResearchEvidence retrieval
Source intake, BM25 query, citations, provenance inspection
Evaluated hybrid RAG and chunking remain release gates 05Kynetra CommerceTransactional inventory
Conditional checkout commits stock and order atomically
MVCC and distributed transactions remain release gates Native transaction proof
Commerce does not borrow another database.
A guarded checkout validates inventory, appends the stock update and order insert as one Universal Log record, and keeps its idempotency receipt across restart. This is single-process atomicity, not a distributed-ACID claim.
Read the architecture →