.webp)
How it works
We built a system of intelligent agents that optimize how your data warehouse scales, schedules, and runs queries.
Each agent targets a specific layer: scaling and sizing, workload placement, and query structure. Together, they help your warehouse - and your data engineering team - do more with less.
Autoscaling Agent
Scheduling Agent
Query Agent
Autoscaling agent
How big should my data warehouse be? How should it scale? Are we using the correct settings?
Smarter scaling
Our autoscaling agent goes beyond the default scaling logic that Snowflake offers, by intelligently optimizing cluster usage based on your actual query patterns.
Predictive warehouse management
By analyzing historical query and data warehousing logs, our agent builds models that predict workload spikes and fluctuations, allowing for more precise scaling decisions.
Lower cost, same performance
Our horizontal scaling of your Snowflake data warehouses solves a constrained optimization problem - minimizing costs subject to not impacting latency.
Scheduling Agent
How long is this query going to take to run on a small data warehouse versus a medium warehouse? Would we save money by using a different warehouse that's already on?
Kubernetes for Snowflake
Our scheduling agent analyzes query behavior in real time and routes each job to the most efficient warehouse — like Kubernetes for Snowflake. No more guessing or overpaying for fixed compute.
Smarter warehouse utilization
Instead of static warehouse assignments, our agent shifts workloads to where they’ll run best. This reduces idle time and improves overall warehouse performance.
Even lower cost, still same performance
By optimizing where and how queries run, we help teams hit higher utilization targets and cut costs — all without compromising speed or reliability.
SQL Query Agent
Which queries are driving up my bill? How do I re-write them to make them more cost efficient?
Reimagined query rewriting
Modern query optimizers are analogous to regular compilers. Our SQL Query agent has broader capabilities and can make more optimizations. It works particularly well for things like RM queries, where those are difficult to reason about for humans.
Using Calcite as a parser
We are primarily using Calcite as a parser, adapted for Snowflake SQL.
LLMs without hallucinations
To mitigate hallucinations, we ensure correctness with a non-ML based formal verification system. This checks your input query is mathematically equivalent to your output query.
select NATION.N_NAME as N_NAME, COUNT(1) as "cnt:C_NAME:ok"
from SNOWFLAKE_SAMPLE_DATA.TPCH_SF1000.CUSTOMER CUSTOMER
inner join SNOWFLAKE_SAMPLE_DATA.TPCH_SF1000.NATION NATION
on (CUSTOMER.C_NATIONKEY = NATION.N_NATIONKEY)
where CUSTOMER.C_MKTSEGMENT not
in ('AUTOMOBILE') and CUSTOMER.C_MKTSEGMENT is not null
group by NATION.N_NAME; select t0.N_NAME as N_NAME, SUM(1) as "cnt:C_NAME:ok"
from SNOWFLAKE_SAMPLE_DATA.TPCH_SF1000.CUSTOMER CUSTOMER
inner join (
select CUSTOMER.C_NATIONKEY as C_NATIONKEY, NATION.N_NAME as N_NAME
from SNOWFLAKE_SAMPLE_DATA.TPCH_SF1000.CUSTOMER CUSTOMER
left join SNOWFLAKE_SAMPLE_DATA.TPCH_SF1000.NATION NATION
on (CUSTOMER.C_NATIONKEY = NATION.N_NATIONKEY)
group by 1,2) t0
on (CUSTOMER.C_NATIONKEY = t0.C_NATIONKEY)
where (case when ((CUSTOMER.C_MKTSEGMENT in ('AUTOMOBILE'))
or (CUSTOMER.C_MKTSEGMENT IS NULL)) then FALSE else TRUE end)
group by 1;