How many units of item B-117 can we still promise to customers?
You have 120 units available.
112 units were sold after 06:00. This answer is wrong.
NDS keeps one live copy of your business data. Ask a question in plain English: AI writes the query, NDS runs it as you, and only the rows you may read come back.
Show me today's orders and who placed them.
Sure, let me do that for you. Here's the query I'm going to use to show you the result:
SELECT o.placed_at, c.name AS customer, c.region, o.amount
FROM sales.orders o
JOIN sales.customers c ON c.id = o.customer_id
WHERE o.placed_at >= CURRENT_DATE
ORDER BY o.placed_at DESC;
Ran as finance@example.com
17 rows from today
| placed_at | customer | region | amount |
|---|---|---|---|
| 11:42 | Nordlicht GmbH | EU | 312.00 |
| 11:15 | Blue Harbor Inc | US | 89.50 |
| 10:58 | Sakura Works | APAC | 1,420.00 |
| 10:31 | Atelier Rive | EU | 64.20 |
13 more rows
Reads customers, orders and payments in every region.
Each new use case tends to bring its own store, its own pipeline and its own rules. The data is the same; the copies multiply.
Search by meaning usually means a second database, with its own copy of your data to secure, back up and keep in step.
Every copy needs a job that moves data, watches for changes and needs someone to look after it.
Rules set in one system rarely follow the data into the next, so AI can end up seeing what people cannot.
Hand AI too much and it gets slow, costly and more likely to guess. Hand it the wrong part and it still answers, with confidence. With NDS, a question narrows one copy of your data to the rows it needs, under the rules of the person asking, without copying anything first.
The question: “Show me today's orders and who placed them.”
Some questions are about meaning, not exact values. For those, NDS will find the items closest in meaning to the question, keep only the ones the person asking may read, and hand just those to AI. Your LLM writes the answer from them.
The question: “Which customers complained about late deliveries?”
AI assistants and agents now answer business questions and act on them. Most read a copy made for them: an export, a warehouse table, a document index. That copy is hours old and often lacks the access rules that protect the original. So the AI answers confidently, from the past.
How many units of item B-117 can we still promise to customers?
You have 120 units available.
112 units were sold after 06:00. This answer is wrong.
How many units of item B-117 can we still promise to customers?
You have 8 units available.
The answer matches what the business sees right now.
Capabilities that come with the database, not as extra systems to buy, connect and secure.
Use the AI model you choose: Ollama, Anthropic or any OpenAI-compatible endpoint, on your servers or in the cloud.
Map, filter and preview data from MySQL, DuckDB and files, then run it now or on a schedule. It lands once and stays current.
Store files and documents next to your rows, link them to records and search them by what they contain.
Ask a question in plain English. AI writes the query, shows it to you and runs it with your permissions.
Keep model training data reproducible, so you can return to the exact data a model learned from.
See what changed in a table and when, and look at your data as it was before the change.
Decide once who may read which rows and columns. Apps, reports and AI questions all follow the same rules.
Write your own logic, deploy it beside the data and run it on a schedule.
If the source has changed since the AI's material was prepared, the answer says so instead of staying silent.
Connect your clouds to NDS. Ask once — NDS reads them all as if they were one. AWS, Azure, GCP, your own data centre: connect them live to NDS. Ask in plain English. AI writes the query. NDS reads each connected cloud in place, joins the results with data it stores itself, and answers — nothing copied, every column labelled.
Each name is a tool, technology or source NDS works with. Select one to see what you can do with it.
Select a name to see what you can do with it.
Bring data from the systems that run your business into the NDS Store, next to everything else, under the same rules.
Tell us which system you need first. Product names are trademarks of their owners; listing them does not mean a partnership.
An agent should see what its person may see, and nothing more.
A business question can take an AI agent several steps. The plan is for every step to use the same current data, stay within the access of the person the agent works for, and leave a record you can review. Here, a sales assistant answers one question.
Illustration. The figures are made up for the example.
Search by meaning finds what a question is about, even when the words are different. NDS will keep the vectors and their index inside the database, next to the data they describe, so a new record can be found without a copy step.
A list of numbers that describes what something means. Texts that mean similar things get similar numbers, even when they use different words. Here, each square is one vector: similar shading, similar meaning.
A map of meaning. Instead of comparing a question with every item, it goes straight to the closest neighbours, the dashed circle here, and leaves the rest alone.
Ask a question above. The answer comes back here.
Pick a question, or add a record and ask again.
Illustration with sample data.
No second database to copy into, secure or back up. The same permissions, backups and restores cover both.
When a record or document changes, just its vectors are marked out of date and refreshed in the background. Nothing is copied again.
NDS narrows by what it already knows, such as customer, region or date, then searches by meaning inside that small set.
The search index is kept apart from the vectors, so it can be rebuilt or swapped without recomputing a single vector.
Move to a new embedding model in one switch-over, never a half-old, half-new mix, and keep the old version for audit.
Search results follow the asking person's current permissions, not rules frozen when the index was built.
AI asks a question. NDS hands back a small result that is close in meaning, allowed for the person asking, and current. The idea is the same wherever data meets AI.
Select an industry to see an example question.
Examples show how NDS fits; they are not customer stories.
NDS does not ask you to replace everything on day one.
Connect NDS to your current databases and ERP. Nothing is switched off.
Dashboards and AI assistants read live data from NDS, under your rules.
Move apps onto NDS when you are ready, and retire the copies you no longer need.
Try NDS locally. It ships as a container.
Run it on your own servers, fully in your control.
Run it in the cloud you already use, next to your data.
The same rules apply whether a person, a report or an AI agent is asking.
Traffic between your tools and NDS is encrypted.
Use the single sign-on you already have, and keep users in sync.
Decide who can read and who can write, in plain SQL.
Every request is recorded: who asked, what was read, and when.
Each request records the agent that acted and the person it acted for, not just a shared service account.
Show what the AI was shown, when, and under which permissions and model, for as long as you choose to keep history.
A business should have one version of the truth, for people and for AI. NDS starts with the database, grows into the place where AI finds exactly what it needs, and goes all the way down to memory hardware built for data.
Five ways to get in touch. Each one starts with a short form.
Want to see NDS work on your own data? Ask for a free trial and we'll help you get it running.
See NDS answer questions on live data, and see the small result AI reads.
Build NDS with us. We are looking for a small number of teams with real data and real AI questions.
Offer a product, data or a service that could help NDS or its customers? Tell us about it.
Every company is putting AI to work on its data. NDS is building the database that hands AI exactly what it needs.
Anything else: support@ndsai.io