Vela pricing
Three tiers, from a free sandbox to a billion-vector index with data residency and an SLA.
Pricing you can read
You pay for the number of vectors stored and the queries you run. Creating an index is not billed.
The Discovery tier stays free, with no time limit. The Team and Enterprise tiers add multiple namespaces, HNSW tuning, data residency and support. Changing tier requires no reindexing.
Tier by tier
What each tier covers, and what is actually billed.
Discovery
Free sandboxOne index, one million vectors, no credit card and no time limit. Enough to validate an integration, compare embedding models or prototype a RAG pipeline.
Capacity
- 1 index
- 1 million vectors
- Up to 4096 dimensions
- 1 namespace
- Metadata filters
Queries
- k nearest neighbours
- Cosine similarity
- 10 queries per second
- 12 ms p99 latency
- 99.2 % recall
Support material
- Documentation at docs.vela.dev
- Python, TypeScript and Go SDKs
- Community forum
- RAG pipeline examples
- Public release notes
Team
Going to productionSeveral indexes, fifty million vectors and technical support within 24 hours. The tier meant for an application served in production.
Capacity
- 10 indexes
- 50 million vectors
- Up to 4096 dimensions
- Unlimited namespaces
- Adjustable shard count
Queries
- 500 queries per second
- Composed metadata filters
- HNSW parameters exposed
- 12 ms p99 latency
- 99.2 % recall
Operations
- Reindexing without downtime
- Daily backups
- p50/p99 latency metrics
- API keys per environment
- Access logs
Enterprise
On quoteA billion vectors per index, your choice of storage region, a contractual availability commitment and a dedicated technical contact.
Capacity
- 1 billion vectors per index
- Unlimited indexes
- Up to 4096 dimensions
- Dedicated shards
- Deployment across 8 regions
Compliance
- Data residency per index
- Encryption with your own keys
- SSO and granular roles
- Exportable audit logs
- Annual security review
Support
- 99.95 % SLA
- Dedicated technical contact
- 24/7 on-call
- Quarterly architecture review
- Reindexing assistance
What gets billed
Units of measureThe bill rests on three units sampled hourly then aggregated over the month. The Team and Enterprise tiers include a flat allowance on each of them.
Storage
- Number of vectors stored
- Vector dimension
- Metadata size
- Backup copies
- Hourly sampling
Queries
- k-NN queries executed
- Value of k requested
- Metadata filters applied
- Batched queries
- Hourly sampling
Writes
- Inserts and updates
- Vector deletions
- Reindexing runs triggered
- Batch imports
- Hourly sampling
Included everywhere
Whatever the tierSome features do not depend on the tier and are available from the first index you create, free tier included.
API and SDKs
- REST and gRPC APIs
- Python SDK
- TypeScript SDK
- Go SDK
- Admin CLI
Search
- Cosine similarity
- k nearest neighbours
- Metadata filters
- Namespaces
- Multilingual search across 40 languages
Portability
- Vector export
- Metadata export
- Change tier at any time
- No minimum commitment
- Full deletion on request
What you measure
The reference values, identical across all three tiers.
Latency
p99 response time for a k-nearest-neighbour query on a loaded index.
Recall
Recall measured on the reference datasets, with the default HNSW settings.
Capacity
Vectors per index, up to 4096 dimensions, spread across several shards.
Regions
Deployment regions, with data residency chosen when the index is created.
Ready to create your first index?
The Discovery tier is free and commitment-free. For billion-vector workloads, write to contact@vela.dev.