Count's compute layer
Count's compute layer intelligently routes queries between your data warehouse, Count's servers, and users' local machines—reducing warehouse costs while enabling faster, more exploratory analytics.
#Why the compute layer matters
The problem: Every query to your data warehouse costs money and takes time. BI tools can consume 30-80% of warehouse costs. When AI agents get involved (running 15+ queries per question), costs can explode.
The solution: Run queries where it makes most sense. Small and medium queries run locally or on Count's servers. Only large initial queries hit your warehouse. Result: 60%+ of queries never touch your warehouse.
#How it works
Query routing:
- Initial query → Runs on your data warehouse to fetch data
- Further queries (100MB-32GB) → Run on Count's servers using DuckDB
- Small queries → Run directly on user's laptop using DuckDB
What this means:
- Once data is pulled from your warehouse, you can iterate, filter, aggregate, and analyze without additional warehouse queries
- AI agents can run unlimited queries for thorough analysis without cost concerns
- Everyone can explore data freely without worrying about warehouse bills
Benefits:
- Up to 80% reduction in BI tool warehouse costs
- Faster query performance (local compute is faster than warehouse round-trips)
- Enables exploratory, iterative analysis without throttling
- Still leverage warehouse for large queries when needed
#Who benefits
Data teams:
- Iterate freely on analyses without cost concerns
- Work with AI agents that can dig deep without limitations
- Faster performance for exploration and refinement
Business users:
- Explore data without needing to understand warehouse costs
- Use self-service analytics safely and affordably
Finance/platform teams:
- Reduce data warehouse spend
- Enable broader data access without cost explosion
- Predictable costs even with increased usage
#Learn more
#Understanding the compute layer
- Query execution model - How Count runs queries
- DuckDB in Count - Local compute engine
- Performance optimization - Writing efficient queries
#Using the Compute layer with Count Metrics
- Count Metrics overview - Governed data models
- Query performance & optimisations - When data is cached vs. live
Last updated: 05/08/26