DataSys builds open-source data management and processing systems for continuously evolving data.

Our work spans stream processing, dynamic graph systems, vector search and indexing, online data maintenance, and reproducible data-system benchmarks.

Website: datasys.sage.org.ai

• IntelliStream incubates early research ideas.
• DataSys develops framework-neutral data systems and benchmarks.
• SAGE uses these capabilities in agent, RAG, workflow, and service orchestration.
• vLLM-HUST develops model execution, scheduling, and hardware acceleration.

• MorphStream - transactional stream-processing engine for ACID transactions over streaming data.
• BriskFlow - vector-native stream processing engine for join-backed semantic windows and continuously evolving data.

• Complete publication timeline - papers, demos, authors, DOI records, official acceptance evidence, and corresponding DataSys repositories.
• CANDOR-Bench - SIGMOD 2026 continuous ANNS benchmark for dynamic open-world streams.
• StreamFP - WWW 2026 fingerprint-guided data selection for efficient stream learning.
• GRACE - accepted ICDE 2026 work on reconstruction cost in dynamic graph processing.
• BriskSnapshot - join-backed semantic windows demonstration reported by its authors as accepted to the ICPP 2026 demo track; the public conference record is pending.

• GRACE - dynamic graph processing for continuously evolving graph data.

• CANDOR-Bench - continuous ANNS evaluation under dynamic, open-world streams.
• vasg - DataSys-maintained VSAG derivative used by CANDOR-Bench for reproducible vector-index evaluation.

• Sesame - data-stream clustering system and benchmark.

• AMM-Algorithms - framework-neutral C++ implementations of approximate matrix multiplication algorithms with Python bindings.
• LibAMM - reproducible benchmark suite for approximate matrix multiplication systems.

• AllianceDB
• CStream
• PECJ
• PDSC
• SentiStream
• StreamFP
• SRTFD

• CFInjectBench
• CC-bench-tools
• StreamProcessing Reading List
• Parallel and Distributed State Management Survey

These repositories preserve upstream algorithms and benchmark baselines used in DataSys evaluation. DataSys ownership does not replace or imply authorship of the upstream projects.

• Parlay-HNSW - ParlayANN-derived HNSW concurrency baseline.
• Concurrent-HNSW - concurrent HNSW evaluation baseline in the hnswlib fork network.
• hnswlib - pinned hnswlib baseline snapshot.
• cufe - DiskANN-network research fork used by CANDOR-Bench.
• IP-DiskANN - pinned CANDOR-Bench IP-DiskANN baseline.
• big-ann-benchmarks - billion-scale ANNS benchmark fork.

Projects graduate from IntelliStream when they have a clear public abstraction, named maintainers, reproducible evaluation, documentation, tests, licensing, and a sustainable release path.

DataSys is not a catch-all state-management umbrella. It owns framework-neutral data systems, indexes, benchmarks, and data lifecycle infrastructure.

See our contribution guide, security policy, and support guide.