The data your AI needs, maintained by the people who know it
Business teams maintain the mappings, classifications and thresholds your models and agents run on, checked at entry and stored in your data platform.
On this page
The context only your business teams can provide
What agents consume at inference time
Cross-table mapping rules, business-defined labels and thresholds, lookup references that connect datasets, and structured decision inputs that tell an agent how to interpret and act on data, so it works from your rules.
What models consume for training and evaluation
Threshold definitions, scoring weights, simulation parameters, calibration values and reference hierarchies, maintained as tables rather than assumptions buried in a notebook, so models train on current inputs.
What agents and models read, and who maintains it
Swipe the table sideways to read each row in full.
| Kind of context | Read by | Maintained by, for example | Used by, for example |
|---|---|---|---|
| Cross-table mapping rules | AI agents | Finance controlling | An agent assigning postings to reporting lines |
| Labels and thresholds | AI agents | Sustainability | An agent flagging plants above their emission threshold |
| Lookup references | AI agents | HR | An agent matching employees to cost centers |
| Decision inputs | AI agents | Marketing operations | An agent choosing which campaign a lead joins |
| Scoring weights | ML models | Marketing | A lead scoring model |
| Simulation parameters | ML models | Sustainability | Emission scenarios per plant |
| Calibration values | ML models | Quality management | A model predicting scrap rates |
| Reference hierarchies | ML models | Finance controlling | Margin models by product group |
Maintained where AI already reads
The context tables live in SAP Business Data Cloud, Databricks or PostgreSQL, next to the data they describe. A business user maintains them in the grid or a form, with value help against your master data, and the change is available to every model and dashboard at once. The master data can live in the same platform or in another connected one.
Trusted by design
Validated at entry
Checked against your master data and your rules before it is written, so a model trains only on values that passed the check.
Authorized down to the row
Roles scoped to folders and row-level rules decide who may maintain which context, the same way they decide it for every other table.
On the record
Who changed which value, and when, sits in the audit log inside your platform, so a model's inputs can be traced to a person and a date.
Platform-resident
The tables live where your models read, in place and as the only copy, and a change is visible at once.
Maintain data and create tables from your AI assistant
New ESG figures from a mail, with the typo caught
A sustainability manager asks the assistant to bring a plant's monthly energy figures from an email into the ESG table. The import preview shows every row before anything is written, and the assistant proposes the fix for the one NextTables refused.
Here is the preview:
| Subsidiary | Month | Source | Figure |
|---|---|---|---|
| BN4230 | Oct 2026 | Electricity | 238,400 kWh |
| BN4230 | Oct 2026 | Heating | 41,250 kWh |
| BN4230 | Oct 2026 | Natural gas | 96,800 kWh |
| BN4320 | Oct 2026 | Diesel (fleet) | 3,150 liters |
NextTables refused the last row: Value (BN4320) does not exist in master data. The other three rows from the same email use BN4230, the key of your Mexico subsidiary, so two digits look swapped. Shall I correct it to BN4230 and import all four rows?
The four October rows for the Mexico subsidiary are in the table and show up in every report that reads it.
A new mapping table from one sentence
A category manager describes the table in a sentence, and the assistant creates it on your platform with its fields and key, ready for the team to maintain.
Product group mapping is ready in your SAP Datasphere space:
- Product, text, part of the key
- Product group, text
- Valid from, date, part of the key
- Comment, text
Your team can maintain it in NextTables right away.
Exactly the permissions of the person
The assistant connects with a personal access token of the person who sets it up, so it reaches exactly the folders, tables and rows that person may reach, and creates tables only where that person may create them.
What the assistant reads
Rows the assistant reads go to the AI assistant you connect, so your AI policy decides which assistant your teams may use. Everything it writes lands in your platform.
A governed gateway for AI agents
Coming Soon
Today an assistant acts as the person who connected it. Agents that run on their own need an identity and limits of their own, and NextTables is designed to serve as their gateway: agent-driven updates follow the same validation, authorization and audit log as a person's, and where your rules require it they wait for a human to approve.Same rules as a person
An agent's write goes through the same validation at entry and the same roles and row-level rules a human user is subject to.
Human in the loop where it matters
Which agent actions pass automatically and which need a person to review and confirm is a decision your team makes, per action.
Every agent change on the record
Agent-driven changes land in the audit log like every other change, so an agent becomes a traceable channel for data maintenance.
Frequently asked questions
The context data AI depends on, still waiting for an owner end to end: semantic mapping tables for agents, classification and labeling tables for models, entity resolution references, threshold and rule tables, and the metadata that grounds a prompt.
Everything on this page except the agent gateway is how NextTables works today, and the value for AI comes from data that is maintained, validated and on the record.
In your enterprise data platform, in the schema your administrator scoped. NextTables keeps configuration metadata only, so models and agents read the platform itself.
The assistant works with exactly the permissions of the person who connected it. NextTables stores or refuses each row on its own and tells the assistant why, and a file goes through an import preview before it is committed. Rolling a change back is on the roadmap.
They are on the product roadmap and marked as such wherever they appear. Release notes and product updates go out by email; sign up on the product updates page.
Bring the table your model is waiting for
Take the deck to the team first, or spend half an hour with us on the mappings, thresholds and references your AI initiative still keeps in a spreadsheet.