GPU-accelerated. Distributed. AI-native.
Orchestrate your AI pipelines with blazing fast SQL.
1# Import required libraries
2from theseus.cluster import GatewayWorkerCluster
3import ibis
4
5# Initialize a Theseus cluster with 1 worker
6tc = GatewayWorkerCluster(n_workers=1).__enter__()
7
8# Connect to Theseus using Ibis and the cluster's location
9con = ibis.theseus.connect(tc._location)
10
11# Read a Parquet file into Theseus
12con.read_parquet('example.parquet', table_name='example')
13
14# Execute a SQL query and convert results to pandas DataFrame
15con.sql('select * from example limit 10').to_pandas()(lower is faster, to the left is cheaper)
By rethinking query execution for GPUs, Theseus parallelizes complex queries and outperforms leading engines, delivering results in seconds.
How it worksCheaper
Faster
Fewer Nodes
Process data and drive LLM workflows on the same GPU. Theseus adds GPU‑powered UDFs to SQL, enabling full pipelines, from embedding and vector search to RAG in Dataframes and SQL.
Check Out RAG UDFs
Theseus provides fast, in-place queries for data lakes, lakehouses, and warehouses, and supports Apache Iceberg and standard file formats.
Read the EBookPowered by a Kubernetes-native control plane, Theseus deploys across public and private clouds using existing authentication and air-gapped support for maximum security and compliance.
Learn more about Theseus Control Plane
Built for heterogeneous, accelerator-native environments, Theseus Query Profiler captures structured execution DAGs at runtime, enabling real-time analysis of query behavior at CPU, GPU, and I/O layers with minimal system impact.
Learn more about the Query Profiler
Run locally. Conda install Theseus and start querying your data immediately.
conda install --override-channels \
-c voltrondata-nightly -c rapidsai -c conda-forge \
theseus-dev quenta cuda-version-12.8 python=3.10Evaluate performance on larger datasets using our cloud-based testing environment.
Deploy Theseus in production with enterprise scale, white-glove support, and custom deployment options.