Tentris: A high-performance, disk-based RDF graph database

Tentris, a high-performance, disk-based RDF graph database developed by DICE Research, claims to reduce RAM consumption from over 10 GBs to MBs and execute analytical workloads that traditionally take minutes or hours in seconds or a few minutes, using worst-case optimal multi-way joins and a compressed Hypertrie index. The database is positioned for AI agents and knowledge graphs, with a live demo available on DBPedia data.

Query more. Wait less. Be the first to try TentrisDB View documentation → how it works One universal index. No tuning. Instead of maintaining multiple indices, Tentris stores RDF graphs in a single compressed Hypertrie that eliminates structural redundancies found in graphs to minimize memory usage. Adaptive query Planning Worst-Case Optimal Joins no vendor lock-in Don't take the claim. Take the benchmark. WDBENCH XAI WatDiv Footprint Built for Data. Designed for AI. Store, query, and reason over massive RDF knowledge graphs with dramatically lower memory and faster execution. Any scale BUILT FOR AI Already using RDF? Install Tentris Download Your first RDF graph in minutes. Same format. No conversion. get started https://docs.tentris.io/quick start.html No setup. Wait less. Query more. live demo https://dbpedia.data.dice-research.org/ui SPARQL 101 tutorial coming soon waitlist The knowledge layer for AI agents. Book a Discovery Call https://calendar.app.google/8sFfnUoJ1dGgJkUBA Simple now. Simple later. For developers, students and researchers Free For data sovereignty and strict compliance Annual For easy setup and enterprise scaling Pay per Use Frequently Asked Questions RDF knowledge graphs is already in use by major organizations across sectors: - cyan Big Tech: /cyan Google uses RDF principles in its Knowledge Graph; Microsoft uses RDF and linked data in its Azure Data Catalog and compliance tools. - cyan Government: /cyan Agencies like NASA, NIH, and FDA use RDF for metadata management, research linkage, and traceability. - cyan Publishing & Media: /cyan Thomson Reuters and the BBC use RDF to interlink editorial and legal content. - cyan Pharma & Life Sciences: /cyan Companies like AstraZeneca use RDF to link biomedical data for drug discovery. - cyan Industrial Engineering: /cyan Siemens, Bosch and Boeing use RDF for product knowledge management and digital twins. RDF is especially attractive to organizations that need long-term data resilience, cross-system knowledge sharing, and intelligent automation. Tentris significantly reduces RAM consumption by executing analytical graph queries in a fundamentally different way than traditional graph databases. Instead of relying on chains of binary joins that materialize large intermediate results either in memory or on disk, Tentris uses pink2 worst-case optimal multi-way joins /pink2 that process all query relations simultaneously. This avoids memory-intensive intermediate states, keeps memory usage predictable, and enables efficient evaluation of complex, highly connected graph patterns cyan from more than 10 GBs to MBs . /cyan At the same time, this algorithmic approach delivers substantially cyan higher query /cyan performance. Worst-case optimal multi-way join algorithms achieve asymptotically optimal runtimes for complex queries—such as triangle or cyclic patterns, which frequently occur in agent-based reasoning—by eliminating redundant computation and efficiently constraining the search space. Combined with a monolithic, highly compressed pink2 Hypertrie index /pink2 and a streaming execution model enabled by multi-way joins, Tentris executes analytical workloads that traditionally take minutes or hours in seconds or a few minutes, while using significantly less memory and compute resources. pink2 Worst-Case Optimal Joins WCOJs /pink2 are advanced database algorithms that compute complex joins faster and more efficiently by processing multiple datasets simultaneously and not just pairwise. This is especially valuable in RDF graph databases, where queries often involve many interconnected triples. RDF graph dbs store data as triples subject-predicate-object , and answering even a simple question like "Who authored a book published by Publisher X?" often requires joining many triples. Traditional pairwise joins can generate huge intermediate results that slow down performance. cyan WCOJs avoid this by computing the join as a whole, /cyan avoiding unnecessary work. Let’s say your RDF graph contains the following triple patterns: A SPARQL query might want to find all ?author who wrote a pink2 ?book /pink2 published by teal PublisherX /teal and in the blue ScienceFiction /blue genre. - A traditional engine would join ?author– pink2 ?book /pink2 , then the result with teal ?publisher /teal , and then with blue ?genre /blue . Each step might create large intermediate sets, even if few results match all criteria. - A WCOJ-based engine evaluates all three conditions together — finding only the matches that satisfy all conditions at once, without bloated intermediate steps. An pink2 RDF Resource Description Framework /pink2 graph database is a type of database that stores data as triples: subject, predicate, object . These triples represent facts, much like simple sentences, and when many are connected together, they form a knowledge graph. For example: teal <Apple /teal pink2 <hasFounder /pink2 blue <Steve Jobs /blue is a triple expressing a factual relationship. Unlike traditional databases, which use rigid tables and schemas, RDF graph databases model data in a flexible, schema-light graph format that reflects the interconnected nature of real-world information. Key differences include: - cyan Schema flexibility: /cyan RDF allows schema evolution without disruption. - cyan Data integration: /cyan RDF excels at integrating disparate and heterogeneously structured data. - cyan Relationship modeling: /cyan Relationships are first-class citizens, unlike in SQL, where JOINs are costly and complex. - cyan Inference and reasoning: /cyan RDF supports automated logical inference based on ontologies and rules. By doing that, it enables machines to deduce new knowledge from existing facts. RDF graph databases are especially well-suited to domains where data relationships are as important as the data itself such as AI, compliance, biomedical research, and enterprise systems. The Hypertrie stores only existing triples using a hash-identifier-based representation for subjects, predicates, and objects. This means that on cyan INSERT /cyan or teal DELETE /teal , only the affected paths and nodes of the index need to be updated. pink2 More specifically: /pink2 - New RDF terms are mapped to compact identifiers. - The corresponding triple is inserted or removed by updating only the relevant branches in the Hypertrie. - Shared prefixes remain unchanged, so unaffected parts of the structure are not touched. This avoids the need for full re-indexing and keeps updates localized. cyan Both. /cyan At core, Tentris is build to excel in complex, analytic SPARQL queries but at the same time support high write throughput. In analytics, we focused for now on accelerating standard SPARQL queries that fail on traditional query engine. Our worst-case optimal join engine mostly eliminates intermediate results and filters down possible solutions rapidly. cyan This allows you to run queries that would otherwise exceed your RAM /cyan or take ages processing circles or large snowflake triple patterns. cyan For updates, we implement MVCC. /cyan So, you update throughput is not disturbed by you colleagues heavy query workload. pink2 RDF Resource Description Framework /pink2 represents data as triples subject-predicate-object using a W3C standard designed for semantic meaning, interoperability, and reasoning. It supports rich ontologies cyan RDFS, OWL /cyan , data validation cyan SHACL /cyan , and global identifiers cyan URIs /cyan , making it ideal for knowledge graphs, linked data, and AI. pink2 Labeled Property Graphs LPGs /pink2 model data as nodes and edges, both with flexible key-value properties. LPGs are widely used in graph databases like Neo4j and are great for real-time analytics, social networks, and operational applications but lack formal semantic standards.