No—DynamoDB cannot perform semantic vector search on raw text alone. Its native vector search compares a query vector with vectors stored in a vector index. You can keep those vectors and your application records in DynamoDB, without a separate vector database, but you still need to create or obtain vector representations for both indexed content and queries.
What “without embeddings” can mean
An embedding is one common way to turn text into a numerical vector that represents its meaning. DynamoDB’s vector index does not require that the vectors come from a particular embedding service, but it does require vectors: AWS describes the feature as similarity search on vector embeddings stored in table items. AWS’s DynamoDB vector index guide explains the feature, while the SearchVectors API reference defines the vector input used for a search.
So the practical distinction is:
- No separate vector database: possible. DynamoDB can hold operational records, vector representations, and the similarity-search index together.
- No vector representations at all: not for native vector similarity search. DynamoDB does not infer semantic similarity from raw text by itself.
- No text embeddings specifically: potentially, if your application creates or obtains another suitable vector representation. The index still searches vectors, not unprocessed text.
How DynamoDB vector search works
You configure a vector index as part of DynamoDB table management. To search it, your application calls SearchVectors with the table name, active vector index name, a query vector, and a TopK value. The query vector’s dimensionality must match the index configuration; the API allows 1–4,096 elements in the supplied vector, and TopK must be 1–100. Vector elements are 32-bit IEEE-754 floating-point numbers. See the SearchVectors API reference for the current input requirements.
The index performs approximate nearest-neighbor (ANN) retrieval. That means it is designed to find nearby vectors efficiently, rather than to match a text phrase exactly. AWS lists semantic search, retrieval-augmented generation (RAG), recommendations, agent memory, and anomaly or fraud detection among possible uses of vector indexes in its DynamoDB vector index guide.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →#1 Best Overall
What you need to provide
- Vectors for the records you want to retrieve. For semantic text search, these are commonly produced by an embedding model before the vector is stored or indexed.
- A query vector at search time. A text query must be converted into a vector by your application or another service before it can be passed to
SearchVectors. - Matching dimensions. The query vector must have the same number of dimensions as the vector index.
- An index and search design. Choose the distance function and the attributes the application needs returned or used in supported search conditions.
AWS’s LangChain example makes this separation concrete: it uses a DynamoDBVectorStore together with a BedrockEmbeddings function. The embedding function supplies vectors; DynamoDB stores and searches them. The AWS LangChain integration guide also notes that vector-index updates are eventually consistent, so a document written moments ago may not immediately appear in search results, and that results are capped at 100.
How to read vector-search scores
A score is not a universal similarity percentage. Its meaning and direction depend on the distance function configured for the index.
| Distance function | How to interpret results |
|---|---|
| Cosine | Lower scores indicate closer matches. AWS documents a range from 0 for identical vectors to 2 for opposite vectors. |
| Euclidean | Lower distance scores indicate closer matches. |
| Dot product | Higher scores indicate closer matches. |
These score interpretations are specified in the SearchVectors API reference. Do not compare scores across distance functions as if they shared one scale.
Search filters and consistency limits
Search conditions can reference fields in the vector index search schema, but the API’s supported conditions are narrower than general DynamoDB expressions. The API reference says HASH and INLINE_FILTER schema attributes support equality only, and only top-level search-schema attributes can be referenced.
Index updates are eventually consistent, as noted in AWS’s LangChain integration guide. Applications that need to show a newly stored item immediately should account for the possibility that a vector search will not return it at once.
Storage and index planning
Vector storage grows with vector dimensionality, projected attributes, and the number of indexed items. AWS’s storage considerations say that, all else equal, a 1,536-dimension vector uses roughly four times the vector storage of a 384-dimension vector. That is a comparison of vector storage, not a total service-cost estimate.
Rank #4
- Choose the smallest dimensionality that meets your relevance needs.
- Project only attributes your application reads directly from search results.
- For production planning, confirm the current service limits and pricing. AWS’s vector index guide lists a maximum of five vector indexes per table and support for on-demand capacity mode.
When to use a different DynamoDB search path
Vector similarity is not the right tool for every retrieval problem. DynamoDB’s conventional secondary indexes support key-based access patterns; they do not find nearest neighbors by meaning or vector distance. AWS describes the distinction in its secondary indexes guide.
| Your requirement | Relevant approach | What it does |
|---|---|---|
| Find records similar to a query by semantic or other vector representation | DynamoDB vector index with SearchVectors |
Retrieves nearby vectors while the records and vectors can remain in DynamoDB; still requires vectors and ANN-aware index design. |
| Retrieve records by exact keys or key ranges | DynamoDB secondary index with Query or Scan, as appropriate |
Supports key-based access patterns rather than nearest-neighbor similarity. |
| Combine vector retrieval with full-text search, analytics, or hybrid retrieval | Evaluate DynamoDB Zero-ETL integration with OpenSearch | Adds a connected search service for broader search needs; AWS presents this as an option to evaluate, not a universal recommendation. |
AWS discusses the OpenSearch option in its DynamoDB OpenSearch integration guide.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Bottom line for implementation
If “without embeddings” means avoiding a separate vector database, DynamoDB’s native vector index can fit that goal. If it means searching the semantic meaning of text without producing any vector representation, DynamoDB’s native vector search does not do that. You can generate vectors outside DynamoDB and store them with your records, but the application must still supply an appropriately dimensioned query vector to search.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
