For companies with data to license

We represent companies licensing their data to AI labs.

Our pipeline cleans and de-identifies your data and prepares it for RL and evals, then several labs bid for it. Your name stays private until you agree to share it.

Find out what it’s worth How we protect you
$ $ $
Works withThe tools you already use, and anything you can export
XeroSendGridQuickBooksAdobeTeradataDropboxCanvaGoogle PlayCassandraAmazon S3SquarePyTorchTelegramOneDriveConfluenceBitwardenMagentoAirflowTerraformGoogle CloudElasticsearchParquetMixpanelSupersetSplunkVercelJiraEpic GamesBigCommerceSnowflakeServiceNowPrometheusAutodeskDockerLINESwaggerTwitchWorkdayPostHogGoogle AnalyticsGustoLinearSupabaseMLflowZohoSharePointHerokupandasGrafanaBasecampAuth0RetoolAirbyteDuckDBJSONLGodotBlenderRobloxMiroGoogle CalendarOracleOktaJupyterOpenSearchTableauLookerBigQueryPower BIAWSUnityCSVUnreal EngineGoogle SheetsPolarsFirebasePayPalTwilioOdooGmailTensorFlowElasticZoomdbt
ClickHouseZendeskSAPBrazeSentryHadoopAsanaWooCommerceGoogle DocsShopifyGitHubPagerDutyCircleCIBitbucketNew Relicmonday.comVonageRedisPlaidNetlifyMariaDBSQLiteMySQLMetabaseDICOMGoogle DriveApp StoreAirtableLoomGitLabDatadogStripeHugging FaceWeChatFigmaFHIR R4AdyenSteamSparkPlayStationYouTubeTrinoHubSpotIntercomGoogle MeetFivetranZapierPipedriveHelp ScoutOutlookSlackHotjarExcelRStudioJenkinsPostmanKubernetesAppsFlyerHiveCleverTapClickUpTrelloCalendlyWeights & BiasesMongoDBDigitalOceanKafkaCloudflareDiscordHL7Microsoft TeamsDatabricksRazorpayWhatsAppBoxMailchimpNotionGitHub ActionsSQL ServerPostgreSQLSalesforceAndroidNeo4j
Market value

Market value discovery

Data has no list price, so we ask several labs to bid at the same time.

$50K to $10M+Typical license value
  • Offers side by sidePrice, exclusivity and term for each bid.
  • We negotiateOn your behalf, and you decide whether to accept.
$
Process

How it works

STEP 1

Intake

Your team uploads one zip file to your AI tool, which runs the intake.

STEP 2

First read

We tell you what labs will pay for and give you a rough range.

STEP 3

Preparation

We clean a sample for you to approve.

STEP 4

Bids

Labs test the sample under NDA and make offers.

STEP 5

Signing

You sign directly with the lab you choose.

Preparation

We prepare your data

Our pipeline does the work, so your engineers only need to export.

example run
$ traindata run ds-02 --sample
✓ profile     repos, PRs, reviews, errors
✓ clean       duplicate and broken rows out
✓ de-identify names, emails, keys replaced
✓ package     JSONL + data dictionary
✓ leak-check  0 identifiers found
→ sample ready for owner review
01ingestIntake zip, then exports
02profileCounts, schemas and gaps
03cleanDuplicates and broken rows out
04de-identifyPII, PHI, secrets, DICOM headers
05packageFHIR, DICOM, JSONL, agent traces
06leak-checkA final scan of every file
RL and evals

Then we add value

Labs pay more for data they can train and test agents on.

Raw records → Clean data → RL tasks and eval sets
# task spec, example
{
  "task":     "take PR to merge",
  "start":    "repo@pre-fix",
  "tools":    ["shell", "git", "tests"],
  "verifier": "test suite passes",
  "split":    "held-out"
}
Task designEach workflow written up with a starting point, the tools allowed and an end state.
Complete episodesTicket, code change and deploy linked as one case.
VerifiersTests that score an agent’s attempt.
RL environmentsCodebases and games an agent can run in.
Held-out eval setsCases kept out of training so labs can use them as benchmarks.
Expert reviewYour experts can be paid to grade cases.
What we support

Data labs are buying

Internal company data that has never been on the public web.

Code01

Private codebases, reviews and the fixes that followed.

GitJSONL

Agents and computer use02

Long tasks with every tool call, approval and outcome.

ATIFvideo + events

Business records03

CRM, support, docs, tickets and accounting.

CSVMarkdown

Health and imaging04

De-identified records, scans and claims.

FHIR R4DICOM

Games05

Shipped projects, assets and player telemetry.

UnityFBXParquet

Calls and voice06

Recordings and transcripts in many languages.

WAVJSONL
See all data types
Protections

Your protections

We act for the data owner in every deal.

AnonymityLabs learn your name only if you agree.
NDA firstSigned before any sample or figure goes out.
Your usersPersonal data is removed before export.
Eval-only samplesLabs can’t train on them and must delete them if there’s no deal.
OwnershipYou license usage rights and keep the data.
Your choice of buyerYou can turn down any lab.
How we are paidA share of the license fee, paid when you sign and not before.
Buyers

For AI labs

Every dataset comes with a data card and an eval sample under NDA.

  • Catalogue of current datasets
  • Data card and dictionary for each
  • Eval-only samples under NDA
Contact

Find out what your data is worth

Your team uploads our zip file to your AI tool, which fills in the intake. No raw data leaves your systems.

Start the intake

Contact TrainData

Open email app