Architecture
Altertable uses a lakehouse architecture with distributed storage and DuckDB-based query Workers. The platform uses cache layers and elastic Worker capacity to handle mixed analytical workloads.
A visualization layer is built on top of the query engine, providing a user-friendly interface for querying and visualizing data through insights and dashboards.
Humans and AI agents use the same data and query surfaces.
Infrastructure
Altertable operates on high-end servers optimized for analytical workloads, offering:
- High-memory Configurations: Optimized for processing large datasets
- Fast NVMe SSDs: Used for local caching and temporary storage
- Multi-core CPUs: Optimized for parallel query execution
The cache layers are:
- Local SSD Cache: Frequently accessed Parquet files are cached locally on each Worker
- Memory Cache: Hot data and query results are stored in memory for sub-second access
Compute resources scale dynamically to meet workload demands:
- Auto-scaling Workers: Additional Workers are automatically provisioned during periods of high usage
- Load Distribution: Queries are distributed across available Workers
- Elastic Capacity: Worker count increases when workload increases
Self-hosted Workers
For environments where data cannot leave your network, you can run Self-hosted Workers inside your own cloud. The control plane continues to manage orchestration, catalog, and UI, while the data plane — query execution, source connectivity, and object storage access — runs in your VPC. See Workers for deployment details.
Learn More
- Analytical Database: Learn how Workers query your data
- SQL Engine: Understand query processing and optimization
- External Catalogs: Connect external data sources to your lakehouse
- Workers: Run the data plane inside your own cloud with Self-hosted Workers
- Performance: Deep dive into performance characteristics
- Limits: Understand system quotas and constraints