Get started with Vela
Create an account, then a first index. Send your vectors and run a similarity query straight after.
How do you begin?
Four steps to go from an empty account to a first query.
Create your account
Open an account with your work email, then generate an API key.
- Free account
- Scoped API key
- Region chosen from 8
Create an index
Declare the number of dimensions and the similarity measure. Vela provisions the shards.
- Up to 4096 dimensions
- Cosine similarity by default
- Optional namespaces
Send your vectors
Push your embeddings in batches, along with their metadata. The HNSW graph is built continuously.
- Batch inserts
- Filterable metadata
- No manual reindexing
Run a query
Query the index for k nearest neighbours, with or without a filter, and read off the latency.
- k-NN search
- Metadata filter
- 12 ms p99 latency
Before you begin
Four things to line up for a first deployment.
An embedding model
Any provider will do. Note how many dimensions it produces: up to 4096 are supported.
A similarity measure
Cosine similarity by default. Dot product and Euclidean distance stay available per index.
A metadata schema
List the fields you will filter on: strings, numbers, booleans, timestamps. They are declared with the index.
A region
Pick from 8 regions based on your data residency constraints and how close your services are.
First index at no cost
An evaluation index, then usage-based billing.
No commitment, no payment method required to open one
Frequently asked questions
Answers to the questions we get most often about getting started.
Arrays of floating-point numbers, up to 4096 dimensions. The dimension count is fixed when the index is created and must match your model's.
Writes are visible to reads as soon as they are acknowledged. HNSW graph construction carries on in the background without blocking in-flight queries.
The metadata filter is applied during graph traversal. On our validation sets, recall stays at 99.2 % and p99 latency at 12 ms.
In the region you chose when creating the index, from 8 available. Data residency never changes without an action on your side.
Ready to create your first index?
Open an account, send a first batch of vectors and run a similarity query.