Search the site
Browse the documentation, the product pages and the engineering log. The keywords below lead straight to the matching sections.
A semantic search
Site search runs on Vela. Your query is turned into a vector, then compared against the indexed pages.
Results are ranked by similarity rather than exact word matching. A query written in plain language therefore returns the relevant pages, even with no shared term.
Search suggestions
The terms most often used to query the site, grouped by topic.
Embeddings
The format of vectors sent to Vela, accepted dimensions and compatible models.
- Dimensions
- Normalization
- Compatible models
Indexes
Available index structures, build parameters and the trade-off between recall and latency.
- HNSW
- IVF
- Quantization
k-NN
Nearest-neighbour queries: top-k size, distance metrics and score thresholds.
- Top-k
- Cosine distance
- Dot product
RAG
Retrieving context to feed a language model from your own documents.
- Document chunking
- Reranking
- Source citations
Latency
Response-time budgets, percentile measurement and behaviour under load.
- p50 and p99
- Caching
- Batched queries
Filters
Metadata filtering combined with the similarity computation, in a single query.
- Metadata
- Boolean expressions
- Namespaces
How to query the site
Four steps to useful results, and what to do when a query returns nothing.
Phrase the query
A full sentence works better than a single word. The query is turned into a vector before being compared against the index.
Narrow the scope
A filter limits the search to the documentation, the product pages or the engineering log. It is applied before the similarity computation.
Read the score
Every result carries a similarity score between 0 and 1. Above 0.7, the match is generally reliable.
No results
If no page clears the threshold, drop the filters or rephrase the query. The shortcuts below cover the most common requests.
Frequent destinations
The most-visited pages, reachable without going through search.
Getting started: create an index and send your first vectors
docs.vela.dev
Open the pageThe search engine: indexes, similarity queries and filters
Up to billion-vector scale
Open the pageIndexed sections
Internal search covers the following four sections of the site.
Product
How the engine works, index structures and similarity queries.
Documentation
API reference, installation guides and annotated code samples.
Integrations
Connectors, SDKs and supported client libraries.
Engineering log
Technical articles, performance measurements and release notes.
Can't find the answer?
The documentation covers the API, the SDKs and the common use cases. For a specific question, write to support@vela.dev.