Cut Snowflake costs
with a proxy you run.

chukei runs in your infrastructure, serves verified repeated reads, lets idle warehouses sleep, and passes unsafe queries straight through to Snowflake. Use replay when you need proof before rollout.

VALIDATED LAUNCH SURFACE
Snowflake Snowflake JDBC Python connector dbt Docker Helm / Kubernetes

THE PROXY PATH

Stop repeat work.
Leave risky queries alone.

CHUKEI_DECISIONsafe savings checks
find repeat work → save / warn / pass through safe
C REPEAT READReuse a verified result SAVE
S IDLE TIMELet warehouses sleep SAFE
W COSTLY SQLFlag expensive patterns WARN
Plan
  1. Recognize repeated safe reads
  2. Keep risky work on Snowflake
  3. Record every avoided credit
Guards
  • Writes are never cached
  • Cache hits are checked live
  • Fallback is direct Snowflake

Start without traffic

CHECK HISTORY FIRST
chukei relay · sf-prod
SELECT region, SUM(amount) FROM analytics.marts.daily_revenue WHERE dt >= DATEADD(day,-7,CURRENT_DATE) GROUP BY 1
REPEAT WORK FOUND
The same safe dashboard query appears again. Fingerprint b3:9f2a…e7 is eligible.
CACHE HIT VERIFIED
No write touched daily_revenue. chukei serves the checked result. Warehouse stays asleep.
TEAM SAVINGS
The saved work is attached to team=growth, app, user, and query source.
SIGNED PROOF
Avoided spend lands in a signed JSON evidence bundle finance can verify later.
Next query…
+ fail-open · +5ms p99
CHUKEI_CONNECTIONS
ACCOUNT: abc12345.us-east-1
SF Your Snowflakecredentials pass through Connected
CL Validated clientsJDBC · Python · dbt Ready
REWRITE_RULES SESSION: sx-7k2m
expensive pattern → safe action
actionflag or rewrite
reviewdeterministic
riskfail-open
fail-open · evidence-ready
RELAY_CONFIG
RUNSyour cloud
TARGETSnowflake only
SUSPENDsuggest-only first
CACHElocal by default
EVIDENCEsigned JSON
PILOT PATH

Run it in your infra.
Prove it on one workload.

chukei PILOT PATH

Point one Snowflake JDBC, Python connector, or dbt workload at chukei instead of changing every client at once. Credentials stay with the driver and SQL stays the same.

Replay QUERY_HISTORY when you need evidence first, then use the proxy to verify cache hits and keep unsafe decisions on Snowflake.

START THE PILOT
1Download chukei2Deploy in your infra3Point one workload4Verify cache hits5Keep unsafe queries direct6Share signed evidence
chukei — relay — zsh
WHY CHUKEI

Cut Snowflake costs.
Keep the data plane yours.

Built for teams that need cost reduction they can validate before a wide rollout: one Snowflake workload, verified cache hits, signed evidence, and fail-open passthrough.

① DEPLOY

Start with one workload

Point one JDBC, Python, or dbt workload at chukei. SQL and credentials stay with the existing driver, and rollback is changing the hostname back.

hostname-only pilot
② CACHE

Reduce repeated compute

Deterministic repeated reads can return from a verified proxy cache instead of burning warehouse credits on the same dashboard or report again.

live Snowflake checks
③ SUSPEND

Cut idle warehouse burn

Detect idle windows and begin in suggest-only mode. Enforce suspend later only for the warehouses and roles you choose.

suggest first
④ ATTRIBUTE

Show who created the cost

Attach spend and avoided credits to the query source: user, app, team, and dbt metadata where available, without relying on perfect query tags.

showback-ready
⑤ PROOF

Validate before rollout

Replay QUERY_HISTORY and export signed JSON evidence so finance, platform, and security can review the savings case before broader routing.

signed evidence
⑥ OPERATE

Keep production predictable

Run the Apache-2.0 proxy in your environment. Cache misses, parse failures, unsafe SQL, and plugin errors pass through to Snowflake.

fails open · no LLM
0-0%
Snowflake compute bill
target to validate
~0k
queries in live
validation
0
cache mismatches
in the soak
~0ms
measured p99
proxy overhead

Savings vary by workload mix and existing tuning. Use replay against your own QUERY_HISTORY before routing traffic.

PROOF BEFORE CUTOVER

Start from history.
Then route one workload.

chukei replay reads a Snowflake QUERY_HISTORY export before traffic moves. Use it to size the opportunity, then deploy the proxy where verified cache hits and safe suspend decisions can actually cut spend.

$ chukei replay --query-history queries.csv --output projection.json --evidence COPY
PROJECTED_SAVINGS / REPLAY SIGNED
parse covered
cache projected
suspend modelled
rewrite flagged
evidence signed
VERIFY WITHchukei evidence verify
READY Route one workload through the proxy. PILOT GUIDE
WHERE CHUKEI FITS

Compare the operating model.
Not just the dashboard.

Hosted optimizers and dashboards help teams see or plan changes. chukei runs in your path so one Snowflake workload can reduce repeated compute, then prove what changed.

JDBC / Python / dbt chukei proxy
your infrastructure
Snowflake
CHUKEI

Self-hosted proxy

Runs inside your infrastructure and sits directly in the Snowflake query path.

Apache-2.0your VPChostname-only pilot
  • Verified cache hits and suspend controls
  • Replay evidence before wider rollout
COMMERCIAL TOOLS

Hosted optimizers

Hosted platforms for visibility, recommendations, forecasting, and managed optimization.

KeeboEspresso AISELECT.devCapital One SlingshotGreybeamSundeckRevefiSedaiUnravel
  • Strong analysis and tuning workflows
  • Often vendor-hosted with pricing and data tradeoffs
OSS / INTERNAL

Dashboards and SQL packs

Useful for explaining where spend went, but usually outside the production query path.

dbt-snowflake-monitoringhomegrown SQLBI dashboards
  • No transparent proxy cache or suspend path
  • Engineers still change jobs, SQL, or warehouses
BUYER CHECK

The core difference is where the tool runs and whether it can act before a human reads a report.

Data plane Your infrastructure

Vendor SaaS or read-only dashboards

Savings action Cache · suspend · flag/rewrite

Recommendations, reports, or manual changes

Rollout risk Replay first, route one workload

Demo, trial, quote, or build reports

Hot path Fail-open · no LLM

Out-of-path or vendor-specific behavior

Run the proxy.
Pilot one workload first.

Download chukei, deploy it in your infrastructure, and point one low-risk Snowflake workload at the proxy. Replay query history first when you need a savings case.

DESIGN DIRECTIONS

Three ways to tell the comparison.

Option 1 / gateway map

Show chukei as the control point.

Client tools enter one proxy. Safe reads return from cache, unsafe work passes through, and savings evidence leaves a signed trail.

JDBC Python dbt
chukei proxy Your infrastructure cache · suspend · evidence
Snowflake direct fallback
Option 2 / floating proof

Show work collapsing into evidence.

Queries, cache checks, suspend decisions, and savings events move into one signed projection before the team routes more traffic.

dashboard query cache hit verified warehouse idle unsafe SQL direct
signed bundle 15-30% target to validate
Option 3 / capability grid

Show what chukei does that reports cannot.

A compact grid makes the buyer check explicit: whether the tool can sit in the path, act safely, and prove savings.

proxy cache suspend recommend dashboard signed JSON vendor report SQL pack fail-open
Selected path chukei runs in your infrastructure