Skip to main content
Mobius
Advanced

What is a vector database

Also known as: vector store

Definition

A vector database stores embeddings and finds the items closest in meaning to a question, even when the question and the document use different words.

Why it matters

For a small business the point is not to run one: AI tools that answer from your files usually keep one behind the scenes.

Common mistakes

Leaving last year's schedule in the searched folder, so the assistant quotes it with confidence.

Related terms

Worked exampleIllustration

Searching a fictional pilates studio's 40 policy passages

A fictional neighbourhood pilates studio keeps a policy sheet of 40 short passages for its WhatsApp assistant. The passages and results are invented for this example.

  1. Stored

    Each of the 40 passages is stored with its embedding.

    40 passages

  2. The question

    A customer writes "I can't make it tomorrow, what do I do?", and the question becomes numbers too.

  3. Keyword search

    A search for the words "can't make it" finds no passage, because no passage uses them.

    0 of 40

  4. Meaning search

    It returns the three closest: "You can cancel a class up to 24 hours before it starts", "To move a class, reply with the day you prefer" and "If you are more than 10 minutes late, the class counts as used".

    3 of 40

  5. To the assistant

    These three passages go into the assistant's prompt, and it answers from them. That step is RAG.

It finds meaning, not words. The AI tool you use usually keeps one for you.

Illustration

Keyword against meaning

CustomerI can’t make it tomorrow, what do I do?

Keyword search

Looks for the words “can’t make it”

No passage uses these words

0 of 40 passages

Meaning search

Looks for the closest meaning

  1. You can cancel a class up to 24 hours before it starts.
  2. To move a class, reply with the day you prefer.
  3. If you are more than 10 minutes late, the class counts as used.

3 of 40 passages

The closeness bars are drawn for teaching, not computed.

A keyword search looks for the same words. A vector database looks for the same meaning, so it finds the passages a customer means even when no word matches.

Definition

A vector database stores embeddings, the lists of numbers that stand for the meaning of each piece of text, next to the text itself. When a question comes in, the question becomes numbers too, and the database returns the pieces whose numbers are closest, which usually means closest in meaning. That is how an AI search finds the cancellation rule for a customer who wrote "I can't make it tomorrow" and never typed "cancel". It can be a dedicated product or an ordinary database with an extension: pgvector, for example, adds vector similarity search to PostgreSQL.

Why it matters

Knowing what it does explains why an assistant finds the right passage in words the customer never used, and why it finds an outdated passage just as easily when the old file is still there. It also holds copies of your documents, so it needs the same care they do.

Directly related: Embedding, RAG, Database.

Improvement tips

  • Keep the documents it searches current and delete old versions, because it finds old text as easily as new.
  • Give a customer facing assistant only customer facing documents, since anything stored can be found.
  • Test the search with questions in your customers' own words, not in the words of your documents.
  • Ask a vendor where the stored copies of your documents live and who can reach them.

Common mistakes

  • Treating the stored numbers as anonymous, when they can reveal much of what the text says.
  • Building your own vector search for a few dozen passages that a ready tool already handles.
  • Judging the search by one lucky question instead of ten real ones.

Related terms

AI and DataAdvanced

Embedding

An embedding is a list of numbers that stands for the meaning of a text or image, so texts that mean the same land close together, even with different words.

AI and DataAdvanced

RAG

RAG (retrieval-augmented generation) is a method where AI answers from your documents: it first finds the relevant passages, then writes the answer from them.

TechnologyBeginner

Database

A database is an organised collection of information, usually tables of rows and columns, that software can search, update and link, like customers to orders.

AI and DataIntermediate

LLM

An LLM (large language model) is AI trained on masses of text to predict the next piece of a sentence. That is how chat tools write answers, right or wrong.

AI and DataBeginner

Hallucination

An AI hallucination is an answer an AI model makes up and states with confidence, such as an invented price, rule or source, because it predicts likely words.

Quick check

A customer writes "I can't make it tomorrow". Why does a vector search find the cancellation rule when a keyword search finds nothing?

Choose an answer

Frequently asked questions

What is a vector database in simple words?
A store that finds text by meaning. Each piece of text is kept with a list of numbers that stands for what it means, and a question is matched to the pieces whose numbers are closest. Texts that mean the same are found even when the words differ.
Does my small business need a vector database?
Usually not as a separate product. If you build your own AI search over your documents, a developer will use one. If you use a chat tool that answers from your files, it keeps one behind the scenes for you.
What is the difference between a vector database and a regular database?
A regular database finds exact matches: this customer number, this date, this word. A vector database finds the nearest in meaning. Some regular databases can do both: pgvector, for example, adds vector search to PostgreSQL.
How is it different from keyword search?
Keyword search finds only the words typed. A customer who writes "I can't make it tomorrow" gets nothing from a keyword search of a policy sheet that says "cancel". A vector search finds the cancellation rule, because the meaning is close.
Why does our assistant quote an old rule?
Because the old document is still among what it searches, and the search finds old text as easily as new. Remove outdated files and keep one current version of each policy.
Why does the search return related passages that are not the answer?
Closest in meaning is not always the answer. Short, focused passages, one rule each, match better than long pages that mix several topics. Test with ten real customer questions and fix the passages that come back wrong.
How do I choose which documents go in?
Only what the assistant should be allowed to say. For a customer facing assistant that means customer facing documents: policies, schedules, answers to common questions. Keep staff notes, contracts and customer records out.
Does a vector database make the answers correct?
No. It finds passages. The model then writes an answer from them, and it can still misread one or add something. Ask for the source of each answer and check anything that reaches a customer.
Is it safe to put customer documents in a vector database?
Treat it like the documents themselves. It holds copies of the text, and research shows the stored numbers can reveal much of what the text says. Keep personal details out unless they are needed, and control who can reach the store. Israel's Privacy Protection Law, 1981, as amended by Amendment 13, in force since 14 August 2025, governs collecting, keeping and using personal details, including details you paste into or upload to an AI tool. Educational material, not legal or tax advice. Check your own case with a lawyer or an accountant.
Can someone read my documents from the numbers alone?
Partly, yes. A 2023 study recovered 92% of short 32 token texts exactly from their embeddings, and recovered full names from a dataset of clinical notes. The numbers are not a safe way to hide text.
Do I need a developer to set one up?
To build your own, yes. To use one inside a ready tool, no: uploading files to a chat tool's project or a support bot usually sets one up for you.
Does it work with Hebrew?
It works with whatever language its embedding model handles, and some embedding models are built for many languages. Test it with real questions in the language your customers write in before you rely on it.

Sources: pgvector README, "Open-source vector similarity search for Postgres" (github.com/pgvector/pgvector), read 2026-09-25, Lewis, Perez, Piktus et al., "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks", NeurIPS 2020 (arXiv 2005.11401), read 2026-09-25, Morris, Kuleshov, Shmatikov and Rush, "Text Embeddings Reveal (Almost) As Much As Text", EMNLP 2023 (aclanthology.org/2023.emnlp-main.765), read 2026-09-25, Privacy Protection Authority, Amendment 13 questions and answers (gov.il/he/pages/tikun13_qa), read 2026-08-31

Last reviewed: 2026-09-25

I have worked with

What Clients Say

From different fields, at different stages of business

Dan Manto

Eclipse Capital

Real Estate Investment and Finance, USA

Arty McLabin

GameReady

Game Development Education and Outsourcing, International

Anna, Beautician

Anna

Beautician

After I finished a cosmetology course at a leading company, I was confused. I didn't know how to open a business, what to sell from home, or how to bring in clients. Since I reached out to Mobius, everything has changed. We built an organized plan with clear steps and no unnecessary risks. Today I see results, and I'm growing and developing every month.

Mark, Massage therapist, Gan Yavne

Mark

Massage therapist, Gan Yavne

When I finished a massage course, I imagined a different world. I found out it's a hard field, and financial stability isn't easy to find. When I came to Mobius, something changed. We started an organized process and built a business plan. Without Mobius? I probably would have given up. Today I believe in myself and in my own path.

Wondering where AI actually fits your business?

Alex helps Israeli businesses use AI and data where it pays off, and skip it where it does not. The first call is free.

+972 055-248-6151. Better on WhatsApp: I am in meetings most of the day and answer as soon as I am free.