Your data. Your bucket. Any cloud.
ManyVector never stores your vectors — it reads and writes to your own object store using credentials you control. Choose the storage backend that fits your cloud, compliance, and cost requirements.
Cost
10x cheaper than RAM
Object storage costs $0.015–$0.023/GB/month. In-memory vector databases charge RAM prices — orders of magnitude more per GB at scale.
Ownership
Your bucket, your bytes
All HNSW graph objects, inverted index files, and raw vectors live in your bucket. No vendor lock-in. Disconnect ManyVector and the data is still yours.
Durability
11 nines by default
Every major object store provides 99.999999999% durability — far beyond what a self-managed vector cluster can offer without significant operational effort.
Scale
Exabyte capacity, no ops
Object storage scales automatically with your data. No cluster resizing, no disk provisioning — add 100 billion more vectors without touching infrastructure.
Five backends. One API.
Amazon Web Services
AWS S3
Standard and Express One Zone storage classes. Any AWS region. SSE-KMS encryption and IAM role-based access.
$0.023/GB
Standard / month
11 9s
durability
Google Cloud
Google Cloud Storage
Multi-region and dual-region replication. CMEK encryption. Workload Identity Federation for GKE deployments.
$0.020/GB
Standard / month
Multi-region
replication
Cloudflare
Cloudflare R2
Zero egress fees. Globally distributed. S3-compatible API — the same SDK, zero code change.
$0.015/GB
storage / month
$0
egress fees
Microsoft Azure
Azure Blob Storage
LRS, ZRS, and GRS redundancy options. Customer-managed keys. Private Link support.
$0.018/GB
LRS / month
RBAC
access control
S3-compatible
MinIO / Self-hosted
Any S3-compatible object store — MinIO, Ceph, Wasabi, Tigris, Backblaze B2. Air-gapped deployments supported.
S3-compat
API compatibility
Air-gapped
deployment support
Common questions
How does ManyVector connect to my bucket?
You provide credentials in ManyVector's config — IAM role ARNs for AWS, service account keys for GCP, or access key pairs for S3-compatible stores. For Kubernetes deployments, workload identity (IRSA, Workload Identity Federation) is the recommended approach so no long-lived credentials are stored on disk.
Do I need to create the bucket before connecting?
Yes. You provision and own the bucket — ManyVector never creates one on your behalf. This keeps your cloud account structure, naming conventions, and billing attribution exactly as you intend. Once the bucket exists and credentials are configured, ManyVector will write its index objects into it automatically on first use.
Is my data encrypted at rest?
Yes — encryption at rest is handled by your cloud provider using whatever key management you configure. AWS S3 supports SSE-S3 (managed keys) and SSE-KMS (customer-managed keys via AWS KMS). GCS supports CMEK. Azure Blob supports customer-managed keys via Key Vault. ManyVector does not add a second encryption layer — you control the keys directly through your cloud account.
What happens to my data if I stop using ManyVector?
Your data stays exactly where it is — in your bucket, under your credentials. ManyVector writes standard objects using a documented open format. You can list them with the AWS CLI, copy them to another region, archive them to Glacier, or inspect them without running ManyVector at all. No migration step is required and no data is held hostage.
Can I change storage backends or use multiple buckets?
Yes to both. Different namespaces can be configured to point to different buckets — useful for multi-region setups or per-environment isolation (e.g. production on S3, staging on MinIO). Changing backend for an existing namespace requires re-ingestion since index files must live in one backing store, but there is no constraint that all namespaces share the same backend.
Does ManyVector add egress costs on top of my cloud provider's rates?
No. ManyVector does not sit in the data path in a way that adds its own egress fees. You pay your cloud provider's standard egress rates, which you can minimize by co-locating your ManyVector instance in the same region as your bucket. Cloudflare R2 charges zero egress, making it the lowest-cost option for read-heavy workloads.