Algeon
Algeon is an experimental, GPU-accelerated query engine built on Apache DataFusion. It brings NVIDIA cuGraph, cuVS and cuML algorithms into SQL as table functions, allowing graph, vector and machine-learning analysis to be composed with relational SQL in one query and one physical plan.
At a Glance
- Data analysts and AI agents
- Data and ML engineering teams
- Rust/DataFusion application developers
- Organizations running NVIDIA-GPU lakehouse analytics
- +1 more
AI Tools by Algeon
(1)
Algeon
GPU SQL Query Engine for NVIDIA
Discussions
No discussions yet
Be the first to start a discussion about Algeon
Latest News
Algeon 0.1.3 released with native S3 Parquet acceleration, read-only Apache Iceberg tables, cuDF aggregation-ordering fixes and updated Flight SQL server image.
Decision Models in the Query Loop: Model Answers Go Back into the SQL, demonstrating model-guided graph rebuilding, catalog cleanup and entity-fragmentation workflows.
Algeon 0.1.2 released with a standalone public Flight SQL server container image and Parquet table registration from a data directory.
Algeon 0.1.1 released with initial public crates and dependency/documentation updates.
Products & Services
Rust/DataFusion-based single-process query engine that runs supported relational plan fragments on cuDF and exposes graph, vector and ML algorithms as SQL table functions.
Arrow Flight SQL server for remote SQL clients; queues concurrent queries, exposes registered functions in the catalog, returns structured errors, and can register Parquet files/directories as tables.
Embeddable Rust/DataFusion session integration that adds Algeon's GPU execution and algorithm functions to an existing DataFusion session.
Table functions for graph traversal, centrality, community detection, components, similarity, core structure and ForceAtlas2 layout.
Market Position
Algeon positions itself as an experimental, in-process GPU SQL engine that runs beside existing query engines and adds graph, vector and ML operators inside the SQL plan. Its stated alternative to a workflow built from separate SQL, export/ETL, notebooks, GPU libraries and re-imports is one composed plan; practical alternatives include CPU SQL plus DuckDB/igraph-style pipelines, or separate graph/vector databases and ML services. Current constraints versus mature platforms include one GPU per query, no spilling, Parquet-focused scans, experimental APIs and no long-lived indexes or models.
Founding Story
Algeon was created to replace the export/download/notebook/GPU-library/re-import glue-code detour common in graph, vector and ML workflows. Its initial vision is an in-process engine where analysts and AI agents can select working sets with SQL, run algorithms over them, join the results back, and iterate on follow-up questions through the same SQL interface.
Business Model
Revenue Model
The project is publicly available open-source software under the Apache License 2.0, distributed as Rust crates and a public container image; no commercial pricing or paid revenue model was identified in the reviewed sources.
Target Markets
- Data analysts and AI agents
- Data and ML engineering teams
- Rust/DataFusion application developers
- Organizations running NVIDIA-GPU lakehouse analytics
- Teams building agentic data services and Flight SQL backends
- Ad-hoc graph, vector and ML pipelines over lakehouse/Parquet data without export scripts or separate graph/vector databases.
- Analysts and AI agents testing hypotheses with iterative SQL queries, changing seeds, filters and controls.
- Agent backends and data services using a remote Arrow Flight SQL endpoint.
- Join- and aggregation-heavy SQL over Parquet with GPU acceleration.
- Campaign audience selection combining graph affinity, vector lookalikes and pretrained model scores.
- Catalog cleanup, brand-entity resolution and other decision-model-in-the-loop analytical workflows.