Data Oil

Data Oil
Resources

The decisions that matter most deserve more than a demo.

Every claim Data Oil makes about cost, security, performance, and control is documented here — written for executives, technical guides written for engineers all the detail required to make an informed decision.

Understand what Data Oil actually does, what it costs, how it stays secure, and whether it is the right decision for your organisation.

No registration. No form. Just read.

Use cases

Four worked bills, each one a month of a real shape of work

Every line is a published meter times a stated quantity. Open one to read it.

Enterprise sales and marketing

One place for the CRM, the collateral and the call recordings

A sales organisation's data is already in four systems: records in the CRM, collateral in a content store, call recordings with a transcription vendor, and a search box over whichever of them somebody indexed. This is what it costs when all four are one database your sellers ask questions of.

The example

A 150-seller organisation, one month

Who this is

An enterprise sales and marketing organisation: 150 sellers, a marketing team, a CRM of four million records, and a library of collateral, contracts and call recordings.

What they do with it

Sellers search accounts and collateral, ask the assistant to draft and to recall, and marketing publishes in two languages. Every call is transcribed and searchable the same day, and the transcript is a record like any other.

What it replaces

The CRM's own search, a content search subscription, a transcription vendor, a translation vendor and whatever answers questions - on one bill, over one copy of the records, with nothing to keep in step.

The bill

$2,889.28 a month, and every line of it

3.0 million reads, 600,000 searches, 3,000 hours of calls transcribed, 25,000 pages of collateral read, 2 languages published and 60,000 questions answered with citations. Multiply any quantity by the rate beside it and you get the figure in the last column.

What was doneHow much of itRate on the cardUSD
Statements that readdashboards, account lists, record lookups. Each one is also an API request and an operation3.0 million$0.0025 per 1,000$7.50
The same statements, as API requestsevery statement, however it arrived4.0 million$0.006 per 1,000$24.00
The same statements, as operations4.0 million$0.006 per 1,000$24.00
Rows those statements returnedcounted per row, so a list of ten costs a tenth of a list of a hundred40.0 million$0.004 per 1,000$160.00
Statements that wroteactivity, notes, stage changes400,000$0.0079 per 1,000$3.16
Search work beyond the read600,000 searches at 7.5 units each - a search is 5 to 10 by the data it examined4.5 million$0.0145 per 1,000$65.25
Call recordings transcribed$0.24 an hour3,000 hours$0.0667 per 1,000$720.36
Collateral and contracts extracteda page, a slide, a sheet or a section; a document's tables become rows25,000 pages$4.80 per 1,000$120.00
Text handed to the translatorcharged once however many languages you ask for - here 20.25 GiB$300.00 each$75.00
Translated text delivered2 languages at 0.74 times the source in bytes, measured0.37 GiB$2,160.00 each$799.20
Questions to the assistant$0.005 a question all in, at the measured question's shape - the request, 2,591 tokens in and 155 out, against your own model60,000 questions$0.0006 each$36.00
Tokens your model was sent155.5 million tokens$0.0005 per 1,000$77.73
Tokens your model produced9.3 million tokens$0.02 per 1,000$186.00
Data held400 GiB$0.10 each$40.00
Bytes sent out to you120 GiB$0.09 each$10.80
Console audit linesevery act a person or an agent took, kept where finance can read it500,000$0.006 per 1,000$3.00
The Large class, held all monththe reservation: connections, ingestions at once, and how much GPU your work may hold730 hours$0.736 each$537.28
Total for the month$2,889.28

$2,889.28 a month is $19.26 a seller

One seat of a reporting tool is $14.00 to $24.00 a month and holds no data at all. The largest single line above is the call recordings at $720.36 - $0.24 an hour, against $0.26 at the cheapest transcription API on our comparison card, which hands back a text file and nothing that can search it.

Before you take this number anywhere

What moves it, and what is not in it

+What the two languages cost, and why that line is the one to watchOpen

Translation is charged twice on purpose: $300.00 a GiB for the text handed to the translator, once however many languages you ask for, and $2,160.00 a GiB for each gibibyte that comes back. This example translates 0.25 GiB into 2 languages, delivers 0.37 GiB and pays $874.20 for it.

Delivered bytes are measured rather than estimated: Japanese comes back at 0.74 times the source in UTF-8 bytes, because an ideograph is three bytes and carries more than an English word. Translate the collateral, not the CRM. A whole free-text column is a much larger number and a much longer wait; the worked invoice on the pricing page prices that one.

+The call recordings, and what they becomeOpen

3,000 hours at $0.24 an hour is $720.36. What you get back is not a transcript file: it is rows in the same database as the CRM records, indexed for keywords and for meaning on the way in, so what did we promise this account about renewal terms is one statement over the calls, the contracts and the CRM together.

+Every rate used above, as publishedOpen the card
MeterRate, USDWhat it counts
api.requests$0.006 per 1,000calls into the served surface
api.operations$0.006 per 1,000statements those calls carried
data.read$0.0025 per 1,000statements that read records
data.create$0.0079 per 1,000statements that wrote records, one meter each for create, update and delete
data.rows$0.004 per 1,000rows a statement touched, counted per row rather than per statement
data.search$0.0145 per 1,000units of search work beyond the read - 5 to 10 a search, by the data it examined
storage.bytes$0.10 eachwhat your data occupies, for the reporting period
storage.backup-bytes$0.05 eachwhat your backup chains occupy
store.egress-bytes$0.09 eachbytes sent out of the deployment to you
ingest.bytes$0.50 eachsource bytes taken in
ingest.rows$0.0005 per 1,000rows stored by an ingest
ingest.items$0.005 eachitems taken in
ingest.prepare.context$0.0015 per 1,000rows knowledge extraction ran over
media.audio-seconds$0.0667 per 1,000audio transcribed
document.pages$4.80 per 1,000pages, slides, sheets or sections extracted
translate.source-bytes$300.00 eachtext handed to the translator, charged once however many languages
translate.delivered-bytes$2,160.00 eacheach gibibyte delivered, so two languages is two deliveries
nlp.requests$0.0006 eachquestions to the assistant
nlp.tokens.prompt$0.0005 per 1,000tokens your model was sent
nlp.tokens.completion$0.02 per 1,000tokens your model produced
audit.records$0.006 per 1,000console audit lines
compute.instance-hour.small$0.184 eachthe Small class, held for you
compute.instance-hour.medium$0.368 eachthe Medium class, held for you
compute.instance-hour.large$0.736 eachthe Large class, held for you
compute.instance-hour.xlarge$4.328 eachthe XLarge class, held for you - the GPU is yours alone

The whole card is 115 meters, read from the running deployment at GET /api/v2/pricing on 4 October 2026. Your console shows it before you are charged and your statement shows it after.

+What is not in any of these numbersOpen
  • Your model. The assistant calls the endpoint you registered with the key you sealed. What your provider charges for those tokens is on their bill. The nlp.* meters above are what Data Oil charges for carrying the question, retrieving the evidence four ways, reranking it and citing it.
  • A discount. Every figure is the published card at the published rate. No minimum, no seat, no plan to outgrow.
  • Anybody else's free tier. Where the platform is hosted is our cost, not a credit passed to you. Outbound data is $0.09 a GiB whoever owns the infrastructure underneath.
  • Tax. All figures are US dollars, exclusive.

Your numbers, not ours

Send three months of invoices and get this page for your own workload

Everything above is one example with its assumptions written down. Send the last three months from your database, search, vector and AI vendors and we will price the same workload on this card, line by line, and hold the figure for twelve months. Still on free tiers? Send the usage instead, and we will price what you will pay when they end.

Send the invoices Start a proof of concept

A business accounting system

A ledger, the paperwork behind it, and an audit line for every act

An accounting system is a write-heavy database, a pile of scanned paperwork nobody can search, and an auditor who will ask who did what. This is what it costs when the ledger, the invoices and the audit are one database, and the retention is a rate rather than a second product.

The example

A mid-market accounting platform, one month

Who this is

A mid-market accounting platform or a finance department running its own: 2.5 million posted transactions a month, 120,000 pages of invoices and receipts, and seven years of it to keep.

What they do with it

Post and reconcile, read the paperwork behind a line without leaving it, search the ledger by words rather than by code, and ask the assistant what moved and why - with the rows it read cited underneath.

What it replaces

The database, the document-capture vendor, the search index over the archive, and the audit store - and the reconciliation job that kept the four of them agreeing.

The bill

$1,447.15 a month for 2.5 million posted transactions

2.5 million postings, 8.0 million reads, 120,000 pages of invoices and receipts extracted, 900 GiB of retained backup chains and 2.0 million audit lines. Every quantity is stated and every rate is the published one.

What was doneHow much of itRate on the cardUSD
Statements that readtrial balances, reconciliations, statements, drill-downs8.0 million$0.0025 per 1,000$20.00
Statements that wrotepostings, allocations and adjustments, one meter each for create, update and delete2.5 million$0.0079 per 1,000$19.75
Every statement, as an API request10.8 million$0.006 per 1,000$64.80
Every statement, as an operation10.8 million$0.006 per 1,000$64.80
Rows those statements returned60.0 million$0.004 per 1,000$240.00
Search work beyond the read300,000 searches over the ledger and the paperwork at 7.5 units each2.2 million$0.0145 per 1,000$32.62
Invoices and receipts extractedthe tables inside them become rows you can join to the ledger120,000 pages$4.80 per 1,000$576.00
Questions to the assistant$0.005 a question all in, at the measured question's shape - the request, 2,591 tokens in and 155 out, against your own model15,000 questions$0.0006 each$9.00
Tokens your model was sent38.9 million tokens$0.0005 per 1,000$19.43
Tokens your model produced2.3 million tokens$0.02 per 1,000$46.50
Data held250 GiB$0.10 each$25.00
Backup chains heldseven years of retention, at half the rate of live data, verified on the cadence you choose900 GiB$0.05 each$45.00
Bytes sent out to you40 GiB$0.09 each$3.60
Console audit lineswho did what, when, and to which record - the line an auditor asks for2.0 million$0.006 per 1,000$12.00
The Medium class, held all month730 hours$0.368 each$268.64
Total for the month$1,447.15

$1,447.15 a month is $0.0006 a posted transaction

Retention is a rate rather than a product: $0.05 a GiB for the period, half what live data costs, on chains verified on the cadence the policy names rather than on the day somebody needs one. The audit is not a log level either - 2.0 million lines at $0.006 per 1,000 is $12.00, and it is the record of who did what to which record that an auditor asks for.

Before you take this number anywhere

What moves it, and what is not in it

+Reading the paperwork behind a line, without a second vendorOpen

120,000 pages a month at $4.80 per 1,000 is $576.00 - just under half a cent a page, where a page is a page, a slide, a sheet or a section. The tables inside a document become rows, so an invoice's line items join to the ledger in SQL instead of sitting in a PDF somebody has to open.

A document-capture service charges for the extraction and hands back JSON you then have to store, index and join yourself - three more things to build, and a copy of the paperwork to keep in step with the ledger.

+Seven years of it, and getting one backOpen

Backups are nomination-only: nothing is copied unless a policy names it, so a scratch database cannot quietly become a retention bill and a production one cannot quietly go uncopied - a database that has been neither nominated nor declined is reported as a defect. Every chain is verified into a clean room rather than against the working copy, and the verification cadence is chosen with the schedule and charged on the same card.

+Every rate used above, as publishedOpen the card
MeterRate, USDWhat it counts
api.requests$0.006 per 1,000calls into the served surface
api.operations$0.006 per 1,000statements those calls carried
data.read$0.0025 per 1,000statements that read records
data.create$0.0079 per 1,000statements that wrote records, one meter each for create, update and delete
data.rows$0.004 per 1,000rows a statement touched, counted per row rather than per statement
data.search$0.0145 per 1,000units of search work beyond the read - 5 to 10 a search, by the data it examined
storage.bytes$0.10 eachwhat your data occupies, for the reporting period
storage.backup-bytes$0.05 eachwhat your backup chains occupy
store.egress-bytes$0.09 eachbytes sent out of the deployment to you
ingest.bytes$0.50 eachsource bytes taken in
ingest.rows$0.0005 per 1,000rows stored by an ingest
ingest.items$0.005 eachitems taken in
ingest.prepare.context$0.0015 per 1,000rows knowledge extraction ran over
media.audio-seconds$0.0667 per 1,000audio transcribed
document.pages$4.80 per 1,000pages, slides, sheets or sections extracted
translate.source-bytes$300.00 eachtext handed to the translator, charged once however many languages
translate.delivered-bytes$2,160.00 eacheach gibibyte delivered, so two languages is two deliveries
nlp.requests$0.0006 eachquestions to the assistant
nlp.tokens.prompt$0.0005 per 1,000tokens your model was sent
nlp.tokens.completion$0.02 per 1,000tokens your model produced
audit.records$0.006 per 1,000console audit lines
compute.instance-hour.small$0.184 eachthe Small class, held for you
compute.instance-hour.medium$0.368 eachthe Medium class, held for you
compute.instance-hour.large$0.736 eachthe Large class, held for you
compute.instance-hour.xlarge$4.328 eachthe XLarge class, held for you - the GPU is yours alone

The whole card is 115 meters, read from the running deployment at GET /api/v2/pricing on 4 October 2026. Your console shows it before you are charged and your statement shows it after.

+What is not in any of these numbersOpen
  • Your model. The assistant calls the endpoint you registered with the key you sealed. What your provider charges for those tokens is on their bill. The nlp.* meters above are what Data Oil charges for carrying the question, retrieving the evidence four ways, reranking it and citing it.
  • A discount. Every figure is the published card at the published rate. No minimum, no seat, no plan to outgrow.
  • Anybody else's free tier. Where the platform is hosted is our cost, not a credit passed to you. Outbound data is $0.09 a GiB whoever owns the infrastructure underneath.
  • Tax. All figures are US dollars, exclusive.

Your numbers, not ours

Send three months of invoices and get this page for your own workload

Everything above is one example with its assumptions written down. Send the last three months from your database, search, vector and AI vendors and we will price the same workload on this card, line by line, and hold the figure for twelve months. Still on free tiers? Send the usage instead, and we will price what you will pay when they end.

Send the invoices Start a proof of concept

A mobile developer, a database per customer

Eighteen hundred customers, eighteen hundred databases, one bill

A tenanted product pays for isolation twice: once for the cluster each customer's data sits on, and again for the smallest plan every one of those clusters has. Here a database costs what it does, a quiet tenant costs almost nothing, and the fence between two customers is the database boundary.

The example

1,800 customers, 250,000 monthly active users, one month

Who this is

A mobile application sold to businesses: 1,800 customer organisations, 250,000 monthly active users between them, and a database of its own for each customer.

What they do with it

The app reads and writes over HTTP from the handset, searches within the customer's own data, and syncs. A new customer is a new database, created by an API call, placed in the region their contract names.

What it replaces

A managed database per tenant with a monthly floor on each, plus a search product with its own per-record charge, plus the sharding code that decided which cluster a request belonged to.

The bill

$2,515.64 a month across 1,800 databases

40.0 million reads, 12.0 million writes, 3.0 million in-app searches and 600 GiB out to handsets, spread across 1,800 databases with no floor under any of them.

What was doneHow much of itRate on the cardUSD
Statements that readevery screen the application draws40.0 million$0.0025 per 1,000$100.00
Statements that wrote12.0 million$0.0079 per 1,000$94.80
Every statement, as an API request55.0 million$0.006 per 1,000$330.00
Every statement, as an operation55.0 million$0.006 per 1,000$330.00
Rows those statements returned120.0 million$0.004 per 1,000$480.00
Search work beyond the read3,000,000 in-app searches, each one inside one customer's own database22.5 million$0.0145 per 1,000$326.25
Questions to the assistant$0.005 a question all in, at the measured question's shape - the request, 2,591 tokens in and 155 out, against your own model100,000 questions$0.0006 each$60.00
Tokens your model was sent259.1 million tokens$0.0005 per 1,000$129.55
Tokens your model produced15.5 million tokens$0.02 per 1,000$310.00
Data held, all 1,800 databaseswhat the data occupies, not what a plan reserved for it300 GiB$0.10 each$30.00
Bytes sent out to handsetsthe largest single line on this bill, and the one to cache against600 GiB$0.09 each$54.00
Console audit lines400,000$0.006 per 1,000$2.40
The Medium class, held all monthone reservation for the deployment, not one per customer730 hours$0.368 each$268.64
Total for the month$2,515.64

$2,515.64 a month is $1.40 a customer, or $0.0101 a monthly active user

A quiet tenant costs almost nothing, because nothing was reserved for it. The smallest dedicated cluster on our comparison card is $0.08 an hour, $58.40 a month, and a product that gives each customer one pays that floor whatever the customer did. Nobody buys 1,800 of them, of course - they share clusters and write the sharding code instead, and that code is the other half of what this replaces.

Before you take this number anywhere

What moves it, and what is not in it

+A database per customer, and what that actually buysOpen

The fence is the database boundary, not a tenant_id column somebody has to remember in every WHERE clause. A customer's data is placed on the instance their contract names, with residency enforced on every write against that instance's region and a model endpoint assigned per database. Creating the next customer is an API call.

Every database has an owner, and every fence - deletion, export, retention, which model may see it - sits inside that ownership. A database nobody owns is reported as a defect rather than quietly belonging to everyone.

+The egress line, which is the one to engineer againstOpen

600 GiB out at $0.09 a GiB is $54.00, the largest single line on this bill. It is charged on what actually left the deployment, so a client that caches, pages properly and asks for the columns it needs pays less. That is something you control rather than a plan you buy.

There is no allowance netted off it and no provider's free tier behind it. The rate is the rate.

+Every rate used above, as publishedOpen the card
MeterRate, USDWhat it counts
api.requests$0.006 per 1,000calls into the served surface
api.operations$0.006 per 1,000statements those calls carried
data.read$0.0025 per 1,000statements that read records
data.create$0.0079 per 1,000statements that wrote records, one meter each for create, update and delete
data.rows$0.004 per 1,000rows a statement touched, counted per row rather than per statement
data.search$0.0145 per 1,000units of search work beyond the read - 5 to 10 a search, by the data it examined
storage.bytes$0.10 eachwhat your data occupies, for the reporting period
storage.backup-bytes$0.05 eachwhat your backup chains occupy
store.egress-bytes$0.09 eachbytes sent out of the deployment to you
ingest.bytes$0.50 eachsource bytes taken in
ingest.rows$0.0005 per 1,000rows stored by an ingest
ingest.items$0.005 eachitems taken in
ingest.prepare.context$0.0015 per 1,000rows knowledge extraction ran over
media.audio-seconds$0.0667 per 1,000audio transcribed
document.pages$4.80 per 1,000pages, slides, sheets or sections extracted
translate.source-bytes$300.00 eachtext handed to the translator, charged once however many languages
translate.delivered-bytes$2,160.00 eacheach gibibyte delivered, so two languages is two deliveries
nlp.requests$0.0006 eachquestions to the assistant
nlp.tokens.prompt$0.0005 per 1,000tokens your model was sent
nlp.tokens.completion$0.02 per 1,000tokens your model produced
audit.records$0.006 per 1,000console audit lines
compute.instance-hour.small$0.184 eachthe Small class, held for you
compute.instance-hour.medium$0.368 eachthe Medium class, held for you
compute.instance-hour.large$0.736 eachthe Large class, held for you
compute.instance-hour.xlarge$4.328 eachthe XLarge class, held for you - the GPU is yours alone

The whole card is 115 meters, read from the running deployment at GET /api/v2/pricing on 4 October 2026. Your console shows it before you are charged and your statement shows it after.

+What is not in any of these numbersOpen
  • Your model. The assistant calls the endpoint you registered with the key you sealed. What your provider charges for those tokens is on their bill. The nlp.* meters above are what Data Oil charges for carrying the question, retrieving the evidence four ways, reranking it and citing it.
  • A discount. Every figure is the published card at the published rate. No minimum, no seat, no plan to outgrow.
  • Anybody else's free tier. Where the platform is hosted is our cost, not a credit passed to you. Outbound data is $0.09 a GiB whoever owns the infrastructure underneath.
  • Tax. All figures are US dollars, exclusive.

Your numbers, not ours

Send three months of invoices and get this page for your own workload

Everything above is one example with its assumptions written down. Send the last three months from your database, search, vector and AI vendors and we will price the same workload on this card, line by line, and hold the figure for twelve months. Still on free tiers? Send the usage instead, and we will price what you will pay when they end.

Send the invoices Start a proof of concept

Research scientists, DNA

Variants, annotations and the literature, in one place you can ask

Genomics data is large, highly structured and useless apart from the papers that explain it. A group usually ends up with a warehouse for the variants, a graph database for the pathways, a vector index for the literature and a notebook holding the three together. This is what one engine costs instead.

The example

A genomics group or core facility, four terabytes in

Who this is

A research group or a core facility: four terabytes of variant calls, annotations and sample metadata, and the literature that goes with them.

What they do with it

Filter cohorts in SQL, traverse gene and pathway relationships as a graph, find papers by meaning rather than by keyword, and ask questions that cross all three - with the rows and the passages cited underneath the answer.

What it replaces

A warehouse, a graph database, a vector index, a document pipeline and the code that joined the four. The variants, the pathways and the papers are the same records here, so a question can cross them in one statement.

The bill

$5,431.78 a month, after a one-off load of $3,097.22

The load happens once and the month happens every month, so they are two tables rather than one total. 4,000 GiB and 2.0 billion rows go in; 1.5 million statements, 400.0 million rows and 30,000 questions come back out.

What was doneHow much of itRate on the cardUSD
Statements that readcohort filters, joins across annotation tables, graph traversals1.5 million$0.0025 per 1,000$3.75
Statements that wrote200,000$0.0079 per 1,000$1.58
Every statement, as an API request1.9 million$0.006 per 1,000$11.40
Every statement, as an operation1.9 million$0.006 per 1,000$11.40
Rows those statements returnedthe biggest line a research workload has, because a cohort is a lot of rows400.0 million$0.004 per 1,000$1,600.00
Search work beyond the readkeyword and vector search over the literature, on the same index as the data1.5 million$0.0145 per 1,000$21.75
Questions to the assistant$0.005 a question all in, at the measured question's shape - the request, 2,591 tokens in and 155 out, against your own model30,000 questions$0.0006 each$18.00
Tokens your model was sent77.7 million tokens$0.0005 per 1,000$38.87
Tokens your model produced4.7 million tokens$0.02 per 1,000$93.00
Data held4,000 GiB$0.10 each$400.00
Bytes sent out to youresult sets pulled down to a workstation or a cluster800 GiB$0.09 each$72.00
Console audit lines100,000$0.006 per 1,000$0.60
The XLarge class, held all monththe GPU is yours alone at this size, which is what embedding a literature corpus wants730 hours$4.328 each$3,159.44
Every month after the first$5,431.78

$8,529.00 in the first month, $5,431.78 in every month after it

The load is $3,097.22 and is paid once: 4,000 GiB at $0.50 a GiB, 2.0 billion rows at $0.0005 per 1,000, and the literature read and lifted into the graph. Splitting it out is deliberate: one total would put a figure on this page that the group pays once and never again.

Before you take this number anywhere

What moves it, and what is not in it

+The one-off load, line by lineOpen the load
What was doneHow much of itRate on the cardUSD
Source bytes taken inVCF, annotation tables and sample metadata, as the platform meters them4,000 GiB$0.50 each$2,000.00
Rows storedparsed, typed, stored and indexed - the base a row costs however little was asked of it2.0 billion$0.0005 per 1,000$1,000.00
Papers and supplements extractedthe tables inside a supplement become rows, so a published cohort is queryable20,000 pages$4.80 per 1,000$96.00
Knowledge extraction over the literaturegenes, variants and pathways lifted into the graph, a model call per chunk. Ask for it on the papers and not on two billion variant rows, which are already structured800,000 rows$0.0015 per 1,000$1.20
The ingestion itself4 ingestions$0.005 each$0.02
Total, paid once$3,097.22

Knowledge extraction is asked for on the papers and not on the variant rows. Two billion variant rows are already structured; a model call per chunk over them would be the largest number on this page and would find nothing a column does not already say. The papers are where the genes, the variants and the pathways are in prose.

+Why the rows-returned line is the biggest oneOpen

400.0 million rows at $0.004 per 1,000 is $1,600.00, larger than the statements that asked for them. That is the shape of a research workload and the card prices it honestly: a cohort is a lot of rows, and a meter that charged per statement instead would make a SELECT * over a chromosome cost the same as a single lookup.

It is also the line a group can move. Filtering in the statement rather than in the notebook, and asking for the columns the analysis uses, is the difference between $1,600.00 and a fraction of it.

+One engine over the variants, the pathways and the papersOpen

The same records carry SQL, full-text, vectors and a graph. A cohort filter is SQL, a pathway walk is a traversal over vertex and edge collections, and finding the paper that explains a variant is a vector search - in one statement, against one copy, inside one transaction. There is no index to sync and no second product's permissions to keep in step with the first.

Residency is enforced on every write against the instance's region, which is what an ethics approval naming a jurisdiction actually requires.

+Every rate used above, as publishedOpen the card
MeterRate, USDWhat it counts
api.requests$0.006 per 1,000calls into the served surface
api.operations$0.006 per 1,000statements those calls carried
data.read$0.0025 per 1,000statements that read records
data.create$0.0079 per 1,000statements that wrote records, one meter each for create, update and delete
data.rows$0.004 per 1,000rows a statement touched, counted per row rather than per statement
data.search$0.0145 per 1,000units of search work beyond the read - 5 to 10 a search, by the data it examined
storage.bytes$0.10 eachwhat your data occupies, for the reporting period
storage.backup-bytes$0.05 eachwhat your backup chains occupy
store.egress-bytes$0.09 eachbytes sent out of the deployment to you
ingest.bytes$0.50 eachsource bytes taken in
ingest.rows$0.0005 per 1,000rows stored by an ingest
ingest.items$0.005 eachitems taken in
ingest.prepare.context$0.0015 per 1,000rows knowledge extraction ran over
media.audio-seconds$0.0667 per 1,000audio transcribed
document.pages$4.80 per 1,000pages, slides, sheets or sections extracted
translate.source-bytes$300.00 eachtext handed to the translator, charged once however many languages
translate.delivered-bytes$2,160.00 eacheach gibibyte delivered, so two languages is two deliveries
nlp.requests$0.0006 eachquestions to the assistant
nlp.tokens.prompt$0.0005 per 1,000tokens your model was sent
nlp.tokens.completion$0.02 per 1,000tokens your model produced
audit.records$0.006 per 1,000console audit lines
compute.instance-hour.small$0.184 eachthe Small class, held for you
compute.instance-hour.medium$0.368 eachthe Medium class, held for you
compute.instance-hour.large$0.736 eachthe Large class, held for you
compute.instance-hour.xlarge$4.328 eachthe XLarge class, held for you - the GPU is yours alone

The whole card is 115 meters, read from the running deployment at GET /api/v2/pricing on 4 October 2026. Your console shows it before you are charged and your statement shows it after.

+What is not in any of these numbersOpen
  • Your model. The assistant calls the endpoint you registered with the key you sealed. What your provider charges for those tokens is on their bill. The nlp.* meters above are what Data Oil charges for carrying the question, retrieving the evidence four ways, reranking it and citing it.
  • A discount. Every figure is the published card at the published rate. No minimum, no seat, no plan to outgrow.
  • Anybody else's free tier. Where the platform is hosted is our cost, not a credit passed to you. Outbound data is $0.09 a GiB whoever owns the infrastructure underneath.
  • Tax. All figures are US dollars, exclusive.

Your numbers, not ours

Send three months of invoices and get this page for your own workload

Everything above is one example with its assumptions written down. Send the last three months from your database, search, vector and AI vendors and we will price the same workload on this card, line by line, and hold the figure for twelve months. Still on free tiers? Send the usage instead, and we will price what you will pay when they end.

Send the invoices Start a proof of concept