Usage and metering

sanda bills on readings it takes itself, not on what you say you will use. A quiet month costs less than a busy one, and you can watch every meter build up in the console before an invoice exists.

The rate card

Prices are in Australian dollars. The pricing page always shows the current rates.

Basic and Standard meter storage and compute at the same rates. Enterprise runs on a plane of its own and has its own storage and compute rates, which the pricing page shows. See Editions.

Storage

Storage is billed in GB-months: how much your warehouses hold, added up across the month. sanda does not bill a month's storage at a single reading. Each reading counts for the time until the next one, so holding more data for part of a month counts for that part of the month and no more.

• What counts. Everything your warehouse holds: the tables you sync or upload, the tables and views you build, and sanda search indexes. An index is storage like any other.
• How it is measured. sanda reads each warehouse's size regularly in the background, and again whenever anyone opens the Overview or Data · Warehouse pages. An owner or admin can also press Measure now on a warehouse.
• What does not count. The sample learning dataset is free and is not metered.

At a warehouse's storage limit, loading pauses. Nothing is billed past the limit. See Limits.

Compute

Compute is billed in compute hours. One compute hour is one vCPU running for an hour. A warehouse scales to zero when nothing is using it, so you pay for the time it is awake and the work it does, and nothing while it sleeps.

• A question wakes the warehouse for a minute. Five questions inside that minute count as one minute. A busy workspace asking small questions all day pays for the minutes the warehouse stayed awake, not for the milliseconds each query ran.
• An awake minute counts at the smallest size, half a vCPU. The work your queries do above that is added on top.
• Everything that uses the warehouse counts. That covers questions from people and from cherry, report blocks, refreshes of the Excel or Power BI feed, scheduled runs, rebuilds of views and procedures, and sanda search indexing.

Each warehouse has a compute limit in hours a day. You are billed the hours used. At the limit, sanda narrows queries rather than stopping them, and pauses loading until use is back under. The hours that still run are billed, so the limit slows spend rather than capping it. You can set a warehouse's limit lower than your edition allows to hold a month down. See Limits.

Managed sync runs

Each successful run of a managed sync is one run on your invoice. A run that fails is not counted. The Managed sync runs line on an invoice shows how many ran and the price per run.

AI

AI is billed by the token, at the rate for the model that answered. Your invoice has one line for each model used, with the number of millions of tokens and the price per million. Which model answers is chosen in cherry. See Models and usage.

Every AI call sanda makes for your workspace counts and is billed, whoever asked for it:

• questions from your team, in cherry and elsewhere;
• sanda's own work on your data, such as learning your schema and suggesting definitions;
• drafts, and sanda search indexing.

Syncs and scheduled reports are not AI and are not counted here. A daily limit caps AI spend, and your owners and admins set it. See Budgets and AI limits.

Where to watch usage

This month so far is the same arithmetic the invoice uses for usage, so the usage you watch during the month is the usage you are billed for at the end of it. It leaves out the flat platform fee. On a trial, the Free trial credit meter shows the credit drawn down. See Free trial.

Estimate a month

The run rate is today's readings carried across a month. It covers storage and compute only, so add the platform fee. It is a guide, not a forecast and not a bill.

Before you have a warehouse, use the estimator on the pricing page. Enter how much data you have, or how many rows, and how many people will use it. It shows the platform fee, storage and compute for one warehouse. Team size affects only how long the warehouse stays awake, since seats are never a line on the bill. Managed sync, AI and sanda search are extra, and depend on what you do with them.

To work it out yourself, add up:

1. the platform fee for your edition;
2. GB-months held, times the storage rate;
3. compute hours, times the compute rate;
4. managed sync runs;
5. AI, which depends on the models you use.