Who has a copy of your pricing table? Nobody knows.
Ask that question in any organisation and watch the room. Somewhere there is a monthly extract going to a distributor, a spreadsheet a consultant still has, a feed to an auditor set up by someone who has left. Every one of those is a copy you cannot revoke, update or audit. OpenSharing is the alternative: publish a share, and the other side queries your live table from where they are, on whatever platform they use — with nothing copied and the whole arrangement visible in one place.
An open protocol, not a Databricks feature.
This distinction decides whether it is usable in practice. Delta Sharing was developed by Databricks with the Linux Foundation as an open, vendor-neutral protocol, and it has since widened into OpenSharing, which Databricks calls the first open, vendor-neutral protocol for securely sharing AI assets.
- 1
You publish a share
A named set of tables, views, volumes, models or notebooks, with a recipient and a grant — all recorded in Unity Catalog.
- 2
They get credentials
Not a file, not an endpoint to build against. A recipient identity under your control, which you can revoke.
- 3
They query it where they are
With SQL, Python or a BI tool, from Delta Lake and Parquet data, with no proprietary client in between.
- 4
They always see current data
Recipients read the latest published version directly from you. There is no refresh schedule to argue about.
- 5
You can see and stop it
Usage is tracked and audited in one place, and revoking the share ends access — with no stale copy left behind.
The recipient does not need to be on Databricks
That is the whole point of a vendor-neutral protocol, and it is what makes this usable with distributors, auditors and partners who have their own stack. Recipients who are not Databricks customers can also reach shared data products through the Marketplace without an account at all.
Databricks names Atlassian, Shell, Nasdaq, Deutsche Börse, SafeGraph and YipitData among the organisations using it.
Not only tables, and that is the recent change.
Sharing started with tables, which was already useful. What has widened is the range of things that can move this way — because an AI asset is as much a thing you might hand to a partner as a dataset is.
Databricks lists structured datasets, unstructured data through volume sharing, AI models, notebooks and code, and under OpenSharing extends the approach to agent skills and unstructured content. Cross-platform view sharing, and sharing of materialized views and streaming tables, are offered as early access — which is worth noting, because “early access” is not the same as a feature you can plan a contract around.
A smaller detail with real commercial weight: sharing through Cloudflare R2 is offered as a way to lower cost, with zero egress fees. If you share large volumes regularly, egress is often the line item that kills the business case, so it is worth pricing both ways.
Delta Lake and Parquet, read live by the recipient from wherever they work.
Unstructured content — documents, images, audio — governed the same way as the tables.
What OpenSharing added: the AI assets, not only the data they were built from.
Which is how a data product arrives with instructions rather than as a mystery.
Cross-platform view sharing, and materialized views and streaming tables, by sign-up.
Direct, marketplace, or neither side sees the other.
The same protocol underneath, three quite different commercial arrangements on top. Choosing the wrong one is the usual cause of a stalled data-sharing conversation.
A share to a named recipient
For a known counterparty: a distributor, an auditor, a subsidiary, a franchise. You know who they are, you have a contract, and you want them reading current data rather than last month’s file. This is where almost everyone should start, because it retires existing extracts rather than creating new work.
Published once, discovered by many
An open marketplace where datasets, AI models, dashboards and notebooks can be discovered and evaluated, including by people not on Databricks. For monetisation, or for making a reference dataset broadly available. The distribution mechanics are the easy half; the licensing and support obligations are the project.
Collaboration without disclosure
For work where neither side may see the other’s rows. Both parties bring data into a hosted environment, agreed computations run there, and only the result leaves — nothing replicated, no raw data exposed, and it works across clouds.
The measurement that could not be done before.
Here is the canonical case, and it is worth walking through because it makes the idea concrete. A retailer knows who bought what. A brand knows who saw its campaign. Neither can hand its customer list to the other — commercially unthinkable, and in most jurisdictions unlawful. So the campaign is measured by proxy, badly, for years.
In a clean room both lists go into a hosted environment, an agreed computation runs — how many people who saw the campaign subsequently bought — and that single number comes out. Neither party sees the other’s rows, nothing is replicated, and the computation is agreed in advance rather than discovered afterwards.
Computations run in SQL and Python; R and Scala are listed, and Java is marked as coming. Predefined templates, notebooks and dashboards are offered for common cases, which matters because the slow part of a clean room is usually agreeing what will be computed, not writing it.
It is not only a marketing tool. We have seen the same shape work for a bank and an insurer measuring shared risk, a manufacturer and a logistics provider reconciling damage claims, and two subsidiaries whose data-protection regimes prevent a straight merge.
Start by writing down what already left.
Before any of this is switched on, we run the same exercise with every client, and it is always revealing. List every place company data currently leaves the building. Not the ones in the integration catalogue — all of them.
The scheduled extracts
The nightly CSV, the SFTP drop, the API someone built for a partner in 2021. Each one is a standing copy with no expiry and usually no owner.
The reports people forward
A dashboard export emailed weekly to a distributor is a data share with no governance at all. It is also the hardest one to find, because it looks like ordinary work.
The consultants and former staff
Every spreadsheet handed over during a project still exists somewhere. You cannot recall it; you can only stop creating new ones.
The ones nobody remembers
There are always two or three. They surface in this exercise, which is reason enough to run it even if you never publish a single share.
Then replace them, one at a time
Each becomes a share with a named recipient, a grant, an expiry and an audit trail — and Unity Catalog becomes the one place that answers “who has our pricing table?” Unity Catalog →
A share you can revoke is a relationship. A file you have sent is a loss of control with a friendly covering email.
The sentence we use to explain why this matters to a boardWhere sharing comes up most.
Two cases dominate: group reporting across entities, and receiving data from your ERP vendor.
SAP Business Data Cloud and Databricks
The connector SAP and Databricks built: SAP data products mounted as catalogues in Unity Catalog, Databricks tables exposed back to SAP, and the business meaning carried across.
Read it →GovernanceUnity Catalog
Tables, files, models, agents and dashboards in one permission model, with column-level lineage, attribute-based policies and a gateway that governs what AI is allowed to touch.
Read it →FoundationsSecurity and Architecture
Control plane, compute plane and your own cloud account: what runs where, which certifications Databricks holds, and the network and key controls you can turn on.
Read it →© 2026 Mawkish Technologies. This guide, its diagrams and illustrations are protected by copyright. You’re welcome to learn from it and to quote it with credit and a link. Reproducing it, by hand or through AI, and publishing it as your own is not allowed. Read our copyright notice
What people ask about sharing.
How is this different from sending someone a file?
Nothing is copied. You publish a share; the recipient queries the current version directly from you with SQL, Python or a BI tool, with no proprietary client in between. Revoke the share and access ends — there is no stale extract sitting in somebody's downloads folder, and no second copy to keep in step.
That is the whole argument for it, and it is a strong one once you count how many of your integrations are really a scheduled export.
Does the other side need to be on Databricks?
No. Delta Sharing was developed with the Linux Foundation as an open, vendor-neutral protocol, and data in Delta Lake and Parquet can be shared with other platforms. OpenSharing extends the same approach to AI assets — models, agent skills, unstructured data. Recipients who are not Databricks customers can reach shared data products through the Marketplace without an account.
Databricks names Atlassian, Shell, Nasdaq, Deutsche Börse, SafeGraph and YipitData among the organisations using it.
What is a clean room for?
Work that neither party is allowed to let the other see. Two companies bring data into a hosted environment, run agreed computations in SQL and Python — R and Scala are listed, Java as coming — and take away only the result, with nothing replicated and no raw rows exposed. It works across clouds.
The classic case is a retailer and a brand measuring a campaign against each other's customer lists. The point is that the measurement happens without either list changing hands.
Can we sell data this way?
Databricks names monetisation as a use case: package datasets, models, dashboards and notebooks once and distribute them through the open marketplace. Sharing through Cloudflare R2 is offered as a way to avoid egress fees on the traffic.
The commercial and legal work around that is yours, and it is the larger half of the project. The distribution mechanics are the easy part.
How do we keep track of what we have given away?
Unity Catalog is the control point: shares, recipients, grants, audit and usage tracking all sit in the one place, which is how you can answer "who has our pricing table?" without a spreadsheet.
We would put that question in your governance review on day one, because the honest answer at most companies today is nobody knows.
Let us list what already leaves your building.
We will go through every scheduled extract, forwarded report and partner feed we can find, say which of them a revocable share replaces, and leave you with one register of who has what — with an owner and an expiry against each. Most clients find at least two they had forgotten.
- Every outbound copy, written down
- Extracts replaced by revocable shares
- Recipients on their own platforms, not yours
- A clean room where disclosure is not allowed
- One register, with owners and expiry dates
Databricks, Lakehouse, Lakeflow, Lakebase, Unity Catalog, Genie, Agent Bricks, OpenSharing, Delta Sharing and Lakewatch are trademarks of Databricks, Inc. Apache, Apache Spark, Apache Iceberg and Spark are trademarks of the Apache Software Foundation. Delta Lake and MLflow are projects of the Linux Foundation. SAP, SAP Business Data Cloud and SAP Datasphere are trademarks of SAP SE. PostgreSQL is a trademark of the PostgreSQL Community Association. All other names belong to their owners. Mawkish Technologies is not affiliated with, endorsed by or a reseller for Databricks, Inc. Everything described here is taken from material Databricks publishes; product names, availability and prices change, so check the current documentation before you commit to anything. Screens shown are our own illustrations with sample data, not screenshots.
At Mawkish, the Orswell Pass is how we keep Databricks OpenSharing work checked, owned and answered for, whoever does the build.