One record, three views.
A conversation carries meaning, source text, time, and relationships. TriviumDB binds vector data, JSON documents, and labeled graph edges to one node identity, reducing synchronization and result stitching across separate databases.
Organize retrieval as a path.
RETRIEVAL
Recall by meaning and by keyword
Combine dense vectors with BM25 text search, while QuIVer provides approximate nearest-neighbor acceleration.
RELATIONS
Follow relationships into context
Continue from a retrieved node across labeled, weighted edges to connect people, events, documents, and concepts.
QUERY LANGUAGE
Compose hybrid queries with TQL
One language connects vector retrieval, property filters, graph expansion, set operations, and aggregation.
PERSISTENCE
Keep every view in one lifecycle
Write-ahead logging, transactions, and atomic publication manage vectors, documents, and graph relationships together.
Embed the database and keep data close.
TriviumDB runs in process through Rust, Python, or Node.js. Rom mode stores data in one portable .tdb file; Mmap mode separates vector and document storage for demand mapping.
The same model can organize conversational memory, personal knowledge, RAG document retrieval, and relationship memory for game characters.