Yuki Cortex Control Center
Snowflake Cortex blends every agent, function and model into one credit pool. Yuki takes it apart and shows spend by product, agent, user and role, from data already sitting in your account.
What it does
Yuki reconciles Cortex spend to your actual invoice, then breaks it down by product, agent, user and role.
Some spend genuinely can’t be attributed. REST inference bills under AI_INFERENCE with no credit usage view behind it. Cortex Search bills on index size rather than searches, so there’s no usage to attribute at all.
We show that share as unattributed rather than spreading it across the things we can measure.
Cortex credits cover the model call. They don’t cover the warehouse time the query around it burned.
If you run a dedicated warehouse for your agents, add it in one click and see Cortex credits and warehouse spend in a single view. That’s the number that matches what you pay.
For AI SQL functions, Yuki shows credits per model alongside P50, P95 and P99 response times.
Before you keep paying a premium for a model, you can see whether it’s actually faster. Often it isn’t.
Latency figures are statement wall times. Useful for interactive calls, less meaningful for batch functions over a whole table.
Set a cost threshold, or let Yuki flag when your daily Cortex total runs at twice its trailing 30-day average.
You don’t need to open a dashboard every morning. You need to hear about it when something moves.
“We tried two other tools before Yuki. They were more gimmick than substance. What resonated with Yuki was the architecture: a real layer in the path of every query, routing it to the right warehouse. Now we have the peace of mind to keep building.”
Albert Obeng,
Head of Technology @ AllianceBernstein
“We tried two other tools before Yuki. They were more gimmick than substance. What resonated with Yuki was the architecture: a real layer in the path of every query, routing it to the right warehouse. Now we have the peace of mind to keep building.”
Albert Obeng,
Head of Technology @ AllianceBernstein
Add Yuki from the Snowflake Marketplace in a few clicks. Read-only access to metadata. No query text, no PII, no code changes.
Yuki reads the Cortex usage already in your account, so spend by product, agent, user and role is there the first time you log in.
Add the warehouses your agents run on, set a threshold or anomaly alert, and hear about it in Slack or email when spend moves.
With Yuki, you get fully automated warehouse optimization that works with your existing setup, no code changes, no disruptions, and no extra work for your team. Just connect, activate, and watch your Snowflake costs drop.
Request a Demo
By clicking Submit you’re confirming that you agree with our Terms and Conditions.
With Yuki, you get fully automated warehouse optimization that works with your existing setup, no code changes, no disruptions, and no extra work for your team. Just connect, activate, and watch your Snowflake costs drop.
Request a Demo
By clicking Submit you’re confirming that you agree with our Terms and Conditions.
Daily Cortex spend in credits or dollars, broken down by product, agent, user and role, against your own baseline. It also shows model cost alongside latency for AI SQL functions, and the share of spend Snowflake doesn't attribute to anyone.
All six that meter separately. Cortex Agents, Cortex Search, Cortex Analyst, AI functions and AISQL, CoCo and CoWork. Each reports through its own usage view with its own shape, which is the reason a single number is hard to produce in the first place.
By role, which is the only field in Snowflake that maps a credit to a group of people. If your Cortex work runs under shared or generic roles, the breakdown will be as vague as the roles are. That's worth fixing before you need the number, not after, and no tool can reconstruct it later.
Not yet, and we'd rather say so. It shows you where to look: which agent carries the largest share, whether a premium model is actually faster, and how much of the bill nobody owns. Turning that into specific recommendations is what we're building next.
No, and neither can anyone else. Snowflake doesn't let you tag a CoCo session to a project, so if two teams fund your work, the spend still lands under your name with no way to split it. We hear this more than almost anything else. It's a platform gap, not a product choice.
No. Each Cortex product meters its own usage, so a given credit appears once, in the view belonging to the product that charged it. Yuki attributes it to that product rather than summing views on top of each other. This is the question technical teams ask us most, and it's the right question to ask any vendor showing you a Cortex breakdown.
Cortex credits cover the model call. They don't cover the warehouse time the surrounding query burned, and Snowflake doesn't attach an agent name to that compute.
If you run a dedicated warehouse for your agents, add it in one click and see both lines in a single view. If your agents share a warehouse with everything else, that compute can't be separated by anyone, including us.
Yes. The total reconciles to your Snowflake billing meter, which is the same number your invoice is built from. The breakdown underneath it is where the difficulty lives, not the total.
We show it as unattributed rather than spreading it across the things we can measure.
Some Cortex spend genuinely has no owner in any Snowflake view. Inference called through the REST API bills under AI_INFERENCE with no credit usage view behind it. Part of AI_SERVICES has the same gap. Cortex Search bills on index size rather than on searches, so there is no usage to attribute at all. Hiding that share would make the report look tidier and trust it less.
Spend appears as Snowflake reports it. One component, the query compute figure inside agent usage metadata, runs on a separate attribution pipeline and arrives later than the rest of the row, which means the current day always looks slightly cheaper than it was. We don't smooth that over, because a number that quietly fills in gaps is worse than one that tells you it's still settling.
Account usage metadata. How many credits were consumed, by which product, under which role, at what time.
Not read: query text, table contents, column names, customer records, anything containing PII.
No. Access is read-only. The Control Center doesn't route queries, doesn't modify warehouses, and doesn't sit in the path of anything you run. Installing it changes nothing about how your workloads execute.
No. Everything comes from usage data Snowflake has already been writing to your account. There's nothing to add to your agents, no SDK, no tagging requirement, and no migration.
Install from the Snowflake or AWS Marketplace, grant read access, open the view. Minutes rather than days. Your history is there from the first screen because the data already existed, so there's nothing to backfill or wait for.
It's free during launch. No card, no contract, no sales call required.
You can, and plenty of teams have. The reason most of them regret it is maintenance rather than effort.