datakumi
Menu
Sign inStart your trial
Sources: link every system's IDsNew

The governed home for master data

Spreadsheet-simple editing for finance, product and procurement teams. Delta tables for data teams.

Start your 14-day trialSee the product
The Products dataset in datakumi: a private draft with three changes, one new record flagged because its market isn't an allowed value, and an approval mark on every record

Works with the stack you already run. Reads the tables your pipelines land. Publishes where they're read.

A day in datakumi
productsv42 publishedsap_de_vendorsread: 12 new IDssuppliersduplicate VAT number refusedaccountsmove scheduled for 1 Januarycustomersrun waiting for approvalcost_centres3 changes approvedcrm_accountsread held: would remove 31% of IDssuppliers4 source IDs linkedproductsUK is not an allowed market
Product

One place for every list the business shares.Edit, check, approve and publish, without leaving the grid.

See the product
product_idnamemarketlist_price
P-1001Oat drink 1LNL2.49
P-1002Rye crackersDE3.10
P-1003Sparkling waterUK0.89
3 rows pasted from Excel

Edit like a spreadsheet

Paste from Excel, bulk edit, undo. Drafts stay private until published.

Requiredname
Allowed valuesmarket
No duplicatesVAT number

Rules on every field

Wrong values are refused when they're entered, not found in the report.

Editor
Approver
Owner

Approve by risk

Sensitive changes need more approvers. Nobody approves their own.

Connect to your stack

Delta tables your engines, catalogs and pipelines already read.

productshistoryaudit
v42
v41
v40
Finance approverData team

Every version kept

See who changed what, when and why. Read the data as of any date.

s3://lake/commercial/products
_delta_log/00042.jsonapproved
_delta_log/00041.json2 days ago
_delta_log/00040.jsonlast week
# any engine, any version import polars as pl pl.read_delta(path, version=41)

Publish to your own lake

Each approved change is a new Delta table version in your storage. No copies held elsewhere.

Sources

Every system's IDs, linked to one record.datakumi reads the tables your pipelines land. It never writes to them.

How sources work
sap_de_vendorslakehouse
salesforce_accountslakehouse
suppliersshop_db
SuppliersLink mode
supplierkey
name
vat_numberlink rule
country

Link mode

Link each source ID to a record people maintain. IDs still to link are issues, and people are told when new ones arrive.

Build mode

Build golden records from the sources. Link rules join them and value rules choose each field. Every run is a change request.

  • Read from any layer: bronze, silver or gold.
  • A read that would remove more than 20% of IDs waits for an owner.
  • Published as one mapping per source, and a joined table.
Why datakumi

Every report is only as good as the lists behind it.Today they live in spreadsheets nobody owns.

Before
  • Several copies of the product list, none of them current.
  • Wrong codes found after the report is wrong.
  • Approvals buried in email threads.
With datakumi
  • One approved version, with full history.
  • Invalid values refused when they're entered.
  • Every change approved and logged.
Integrations

Read from your systems. Publish where your data is read.Open formats, in your own storage.

For data teams
ConnectionsRead the tables your pipelines land, from any layer. Never written to.
PostgreSQLSnowflakeBigQueryRedshiftDatabricksTrinoAmazon S3AzureGoogle Cloud StorageMinIO
StorageDelta tables land in your own bucket.
Amazon S3Azure Data LakeGoogle Cloud StorageMinIO
QueryEvery engine that reads Delta reads datakumi.
DatabricksApache SparkMicrosoft FabricTrinoDuckDBPolarspandas
PublishingRegistered, described and tested where your team looks.
Unity CatalogAWS Gluedbt seedsGit
FormatsOpen, so the data is never locked in.
Delta LakeParquetCSV
AI assistantsAny MCP client, with the model you choose.
ClaudeGitHub CopilotOllamaAny MCP client
Sign-in and approvalsYour identity provider, and the tools people already answer.
Microsoft Entra IDOktaOpenID ConnectSlackMicrosoft TeamsEmail
RunOne container next to the database you already operate.
DockerTerraformPostgreSQLPrometheus
AI assistants

Bring your own model.Claude, Copilot or a local model works with your datasets through MCP.

Set it up
Which suppliers have no country, and which look like duplicates?

14 suppliers have no country, and 3 pairs share a VAT number.

I put the fixes in your draft. Nothing is published until it's approved.

Your draft17 changesAI assistant (MCP)Submit for approval
Acts as youYour organisation, datasets, fields and record limits. Nothing more.
Changes only go into your draftIt can submit a change request when you ask. It can't publish, approve or reject.
Read-only if you wantWith a read-only token it answers questions and runs SQL, and changes nothing.
In the audit logEverything it does is logged as you, marked AI assistant (MCP).
Enterprise

Controls IT and auditors ask for.Your cloud, your sign-in, a log of every change.

Platform and security
Your cloudDocker, PostgreSQL and your object storage. No data leaves your account.
Every changeApproved by someone else, and kept in a tamper-evident log.
Open formatsDelta, Parquet and CSV. Your data is never locked in.

Single sign-on

Microsoft Entra ID, Okta or any OpenID Connect provider.

Row- and field-level access

Each market sees its own rows. Sensitive fields stay hidden.

Approve anywhere

In the app, or from email, Microsoft Teams or Slack.

Pricing

From €59/month.Charged per editor. Viewers and approvers included.

See pricing
Per editor

Managed

From €59/mo

You bring your data lake. We run the rest.

  • Charged per editor — viewers and approvers are included
  • Every product feature: rules, approvals, audit and Delta tables
  • Hosted, operated and monitored by us
  • Upgrades, backups and restore checks handled
  • Single sign-on set up with your identity provider
  • Onboarding, and support with agreed response times
Start your 14-day trial
Docs

Everything to use and run it.Guides for business users, for the team that hosts it, and for sizing.

Read the docs

Getting started

Run it, create a dataset and publish it: from nothing to a Delta table in five steps.

Read

User guide

What datakumi does, how business users edit and approve, and how sign-in and access work.

Read

Self-hosting

Run datakumi with Docker: where its data lives, your storage, publishing and configuration.

Read
FAQ

Questions, answered.Anything else? Book 20 minutes.

What is master data?

The shared lists other data refers to: products, customers, suppliers, accounts, locations and code lists.

Who uses datakumi?

Business teams edit and approve the data in the browser. Data teams set the rules and read the results.

Where is the data stored?

As Delta tables in your own Amazon S3, Azure or Google Cloud storage. Settings and approvals are kept in your PostgreSQL database.

Does it replace our ERP?

No. It governs the lists that your ERP, BI tools and pipelines share, and publishes them where they're read.

Does datakumi connect to SAP or Salesforce?

No. Your pipelines land their tables in the lake or a warehouse, and datakumi reads them from there. It never writes to them.

Can we use our own AI assistant?

Yes. datakumi has an MCP server, so Claude, GitHub Copilot or a local model can work with your datasets. It acts as you, and its changes only go into your draft.

How large can a dataset be?

Up to a few hundred thousand records per dataset stays interactive. It's built for master and reference data, not transactions.

Bring one spreadsheet.Leave with a governed dataset.

Book a 20-min demoStart your 14-day trial