# Every data flow, under your control

> Data, Substrait's second pillar. Substrait brokers every flow of data between a company's systems and its AI-built tools: the data owner approves it, Substrait carries it through one gateway, and every read is on record.

URL: https://substrait.build/data/  
Last updated: 2026-11-10

Substrait brokers every flow of data between your systems and your AI-built tools: the data owner approves it, Substrait carries it, and every read is on record.

- **Reach every system through a single gateway.** Your ERP systems, your data warehouse and the tools your people built, and no tool holds a key. [See how](#every-system-through-a-single-gateway)
- **Controlled by your data owners.** Each tool reads only what its data owner approved, for a set time. [See how](#data-owners-control-every-flow)
- **Safer than the usual routes.** No pasted keys, shared logins or spreadsheet exports. [See how](#safer-than-the-usual-routes)
- **Built on the right data.** Builders find reviewed, well-defined data, so tools and dashboards are right. [See how](#builders-find-the-right-data)

## Every system through a single gateway

| Source | How a tool reaches it |
|---|---|
| Your ERP systems | Connectors for ERP systems such as SAP, Oracle, NetSuite and Microsoft Dynamics 365, and for Salesforce, SQL Server, SharePoint and Microsoft 365 |
| Your data warehouse | Reviewed, parameterised and versioned data sets that carry the data dictionary, never raw queries |
| Tools your people built | Each tool's API joins the library automatically |

Every source goes through the same gateway and the same permission model.
Substrait holds every credential and attaches it to each call, so no tool holds a key.

### Your data stays where it is

When Substrait runs in your own cloud, the gateway, the connections and the credentials all sit inside your environment.
Data moves between your systems and your tools without leaving your perimeter.
Substrait's control plane sees the decisions, never the data.

## Data owners control every flow

Reaching a system is the easy part.
On Substrait, the person who owns the data decides which tool may read it, and for how long.

<figure class="art">
  <img src="/illus/art/grant.svg" alt="One grant on Substrait: the Returns tracker may read Orders, read-only, approved by the finance data owner, for 90 days, with every read on record." width="1040" height="380" loading="lazy" />
</figure>

- **The owner approves each tool.** Data is grouped by business domain, such as finance or operations, and each group's owner approves each tool.
- **For a set time.** Every grant expires, then it is requested again.
- **Never further than its builder.** A tool's access cannot be wider than what the person who asked for it is enrolled to see.
- **Every read on record.** Substrait records which tool read what, and the person behind it.
- **Switched off at once.** A withdrawn grant refuses the tool's next call, with no redeploy.

### More controls for your data owners

| Control | What it does |
|---|---|
| **Down to the field** | A data group can name single fields, so one sensitive column stays closed on an otherwise open API |
| **A limit on calls** | When approving a grant, a data owner can cap how often a tool may call, so a runaway tool slows down without affecting others |
| **Rules for sensitive data** | Your organisation can decide which tools may go live at all if they use certain sensitive fields |
| **Tools calling tools** | A call from one tool to another goes through the gateway too, approved by the owner of the tool being called |
| **Nothing goes out unseen** | A tool's outbound traffic is denied unless a grant allows it, so the gateway is the only way out |
| **Check before you grant** | An admin can preview what a set of groups would give a person, before anyone receives it |
| **Requests in one inbox** | Every request for data reaches its owner through Substrait's communications centre, with the reason attached |
| **Agents under the same grant** | An agent's access is never wider than its owner's, and every action is recorded as the agent acting for a named person |

## Safer than the usual routes

Most companies already let tools reach their data in one of a few ways.
Each one leaves a question nobody can answer.

| The usual route | What goes wrong | On Substrait |
|---|---|---|
| A key or password pasted into the tool | Anyone with the code holds the key. Revoking it means finding every copy. | Substrait holds every credential and attaches it to each call. The tool never sees it, and revoking takes effect on the next call. |
| A shared service account | Every tool reads as one user, so nobody knows which tool or person read what | Each tool has its own identity and its own grant. Every read is attributed to the tool and the person behind it. |
| Each person connects their own apps to their AI assistant | Access follows one person's account, and the company cannot see or approve the flow | The data owner approves each flow for the company, and every flow is visible in one place |
| An export into a spreadsheet | The copy goes out of date, spreads by email, and leaves your control | Tools read the live data through the gateway, and nothing is copied out |
| Direct access to the data warehouse | Anyone with access can query anything, with no agreed definitions | Reviewed, parameterised and versioned data sets, each carrying the data dictionary and a limit on use |
| A ticket to IT for each new connection | Slow, built one tool at a time, and the access never expires | The data owner approves a whole data group once, from their inbox, and every grant expires |

## Builders find the right data

Builders browse the API Library in conversation with their AI model.
Data nobody granted you is invisible: no name, no count.
The right data, found early, makes a better tool.

- **The builder's AI picks the right source.** Each entry in the library says what it is for and what it is not for.
- **Dashboards start from agreed definitions.** Warehouse data sets are reviewed and versioned, and each carries the data dictionary, so two teams' dashboards start from the same definition of a figure.
- **Real data while building.** A data owner can let a builder see real responses, read-only and for a set time, so the tool is right before it goes live.
- **Tools build on tools.** Each tool's API joins the library, so the next builder starts from it.
- **Live data, not exports.** A dashboard reads through the gateway, so it does not go stale as a spreadsheet copy does.

<figure>
  <img src="/screenshots/65_api-library-tab@2x.png" alt="Substrait console: the API Library, listing company APIs with their owners and data groups." width="2570" height="509" loading="lazy" />
  <figcaption>The API Library. Product UI. Sample data.</figcaption>
</figure>

## Questions and answers

### Which systems can tools built on Substrait connect to?

Your ERP systems, such as SAP, Oracle, NetSuite and Microsoft Dynamics 365, your data warehouse, and the tools your people built on Substrait. Every source goes through one gateway.

### Who decides which tool may read which data?

The data owner. Each set of company data sits in a data group with a named owner, who approves each tool's access for a set time.

### Does a builder need to know the database password?

No. Substrait holds every credential and attaches it to each call through its gateway. Builders and data owners never handle one.