Object storage, indexes and permissions in one server.Object storage, indexes and permissions in one server.
Run a local process while building your application, then deploy Keldra across your own servers as your storage needs grow. Original files, declared indexes and relationship-based permissions share one platform.
Correctness, scale and performance guide the design
Correctness comes first, followed by scale and performance. Keep control of your infrastructure, place data across suitable failure domains and choose the capacity your workload needs.
Original files, distributed storage and indexed queries.
ARCHITECTURE OVERVIEW
Object APIs, searchable indexes and distributed storage
Applications use object and query APIs. Cluster nodes place and protect the stored data and serve the supported access paths. You do not need a separate search service just to use Keldra's declared indexes.
Original objects
Store JSON, images, video and other files as their original bytes. Large objects are split across storage resources and reconstructed through the object APIs.
Purpose-built indexes
Declare the supported fields, terms or vectors your application queries. Indexes provide access to searchable representations without replacing the original object.
Familiar interfaces
Use supported S3 operations, native clients and the optional Git HTTP gateway. Protocol support has defined boundaries; it does not imply every feature of another platform.
Your applications
Native APIs · supported S3 operations · indexed queries
Keldra cluster
Object storage · declared indexes · permissions
Node 1
Storage and request serving
Node 2
Storage and request serving
Node 3
Storage and request serving
Node 1
Storage and request serving
Node 2
Storage and request serving
Node 3
Storage and request serving
Node 1
Storage and request serving
Node 2
Storage and request serving
Node 3
Storage and request serving
Your infrastructure
On-premises, private cloud or public-cloud servers
A deployment overview, not separate services you must install. Place replicas and erasure-coded data across suitable failure domains for your durability requirements.
ARCHITECTURAL SCALE TARGETS
Scale targets for nodes, buckets and objects
Lightweight buckets, distributed object placement and relationship-based permissions guide Keldra's design for large installations. These are architectural targets across a cluster.
Capability
Design target
Scope
Cluster nodes
1,000
Across the cluster
Users and subjects
Billions
Across the cluster
Buckets
Tens of millions
Across the cluster
Files and objects
Hundreds of billions
Across the cluster
Size a deployment for its object sizes, query mix, redundancy and recovery headroom. Add nodes and capacity as those requirements grow.
Your data. More control. A clear operational plan.
Durable files
Configure replication or erasure coding to meet the workload's durability needs. Capacity planning must include the physical redundancy and recovery headroom, not only logical file sizes.
Bounded atomic programmes
Use explicit deterministic state-transition definitions when several paths must change together. These are bounded operations, not uploaded executable code or a transaction wrapper around every ordinary file upload.
Separate access responsibilities
Application and bucket permissions protect object operations. Custom realms model your product's users and can filter indexed results for a selected subject and relation.
Plan backup and recovery
Replication is not an independent backup. Keep recoverable copies, retain the configuration and credentials needed to restore, and rehearse restore and export with the exact server release before relying on the procedure.
Durable files
Configure replication or erasure coding to meet the workload's durability needs. Capacity planning must include the physical redundancy and recovery headroom, not only logical file sizes.
Bounded atomic programmes
Use explicit deterministic state-transition definitions when several paths must change together. These are bounded operations, not uploaded executable code or a transaction wrapper around every ordinary file upload.
Separate access responsibilities
Application and bucket permissions protect object operations. Custom realms model your product's users and can filter indexed results for a selected subject and relation.
Plan backup and recovery
Replication is not an independent backup. Keep recoverable copies, retain the configuration and credentials needed to restore, and rehearse restore and export with the exact server release before relying on the procedure.
Durable files
Configure replication or erasure coding to meet the workload's durability needs. Capacity planning must include the physical redundancy and recovery headroom, not only logical file sizes.
Bounded atomic programmes
Use explicit deterministic state-transition definitions when several paths must change together. These are bounded operations, not uploaded executable code or a transaction wrapper around every ordinary file upload.
Separate access responsibilities
Application and bucket permissions protect object operations. Custom realms model your product's users and can filter indexed results for a selected subject and relation.
Plan backup and recovery
Replication is not an independent backup. Keep recoverable copies, retain the configuration and credentials needed to restore, and rehearse restore and export with the exact server release before relying on the procedure.
Get started on your terms.
Try Keldra on your own infrastructure, test the workflows your application needs and keep an independent route to your original files.