Vela documentation

Create an index, ingest embeddings and query your vectors by similarity

Everything you need to get going

Create an index, send a first batch of embeddings and run a cosine similarity query in a matter of minutes.

Whether you are building a RAG pipeline, a product search or a recommendation engine, the documentation covers ingestion, metadata filters, namespaces, HNSW tuning and reindexing.

Getting-started guides
Python, TypeScript, Go SDKs
API reference
Release notes

Documentation with structure

Six categories covering the full lifecycle of an index

01

Quickstart

Create an API key, declare an index, send a first batch of embeddings and run a k-NN query.

02

API reference

REST and gRPC endpoints: index, upsert, query, metadata filters, statistics and quotas.

03

SDKs

Python, TypeScript and Go libraries, with batch ingestion, error recovery and result pagination.

04

Tutorials

Create an index, ingest vectors, filter by metadata and wire up a RAG pipeline end to end.

05

Operations

Size your shards, tune HNSW and balance recall against p50/p99 latency under load.

06

Release notes

Version history, API changes, announced deprecations and migration procedures.

12 ms p99 latency

Answers in milliseconds

The HNSW index answers k-nearest-neighbour queries under 12 ms at the p99.

An index takes up to 1 billion vectors and 4096 dimensions. Measured recall reaches 99.2 % on the reference datasets. Metadata filters are applied during graph traversal, with no post-processing.

12 ms p99 latency
99.2 % Measured recall
4096 Max dimensions

Documented use cases

RAG Product search Recommendation Deduplication Multilingual search

SDKs and tooling

The libraries and resources that surround the API

Official SDKs

Three libraries maintained by the team, with full typing, error recovery and batch ingestion. Every example in the documentation is given in all three languages.

  • Python SDK
  • TypeScript SDK
  • Go SDK
  • REST and gRPC clients

Integrations

Connectors for the most common ingestion pipelines and RAG frameworks.

Regions

8 deployment regions, with data residency chosen per index.

Release notes

API changes, deprecations announced ahead of time and versioned migration guides.

Support

Technical questions at contact@vela.dev, plus discussion in the public user community.

Data residency

Vela is deployed across 8 regions. You pick the region when you create the index, and the vectors never leave it.

Default region

eu-west-1
Paris, France

Entry points

API REST on api.vela.dev
gRPC Batch vector ingestion
CLI Index and namespace administration
8 regions available Americas, Europe, Asia-Pacific

A question about the API

The full documentation is published at docs.vela.dev. For anything else, write to contact@vela.dev.