Frequently Asked Questions

VectorScaleDB is a coupling intelligence engine built on a temporal-semantic database. The coupling matrix discovers cross-domain behavioral relationships that no other system can see. It unifies time-series data and vector embeddings in a single proprietary index and enables queries no combination of existing databases can replicate -- like "find entities that behaved like this one did yesterday," "predict which domains will be affected when this regime changes," or "what does this data mean in context of everything else?" Safety is enforced by mathematics, not policy.

Those are separate systems that require custom ETL, application-level joins, and duplicated data. VectorScaleDB provides temporal-semantic queries as native database operations with a unified index, so temporal KNN, trajectory similarity, behavioral compression, and coupling-driven prediction all happen in one engine with sub-millisecond latency. More importantly, the coupling matrix discovers cross-domain relationships between entity types that no combination of single-purpose databases can detect.

REST API (450+ endpoints), gRPC, GraphQL (55+ queries, 9+ mutations, subscriptions), WebSocket, and MCP (30+ tools across 6 categories for AI agent integration). SDKs for Python, Rust, and Node.js. Go SDK planned.

200–16,500x on vector data, with cross-entity behavioral correlation adding 2–5x on correlated clusters. Non-vector data (documents, graphs, blobs) achieves 30–500x via semantic chunk deduplication. Compression is per-entity-type optimized.

Up to 4,096 dimensions by default, configurable via environment variable.

Depends on deployment size. Batch ingestion accepts up to 10,000 vectors per request with configurable parallelism. Multiple concurrent batch requests can achieve aggregate throughput well above 100K vectors/second on appropriate hardware.

36+ native adapters including MCAP, LAS/PCD, SPZ, glTF, NetFlow/sFlow, SNMP, gNMI, Prometheus, FIX protocol, NWB, HDF5, NEV/NSx, SpikeInterface, MEA streams, FlyWire connectome, FlyGM GNN, Brian2/NEST spike trains, MuJoCo behavior, MQTT, BVH/C3D motion capture, ADS-B, AIS, VCF genomic variants, gene expression, DNA foundation model embeddings, BCI EEG signal frames, and more. Any other data can be ingested via the standard vector ingest endpoint.

Yes, the default entity type is GENERIC. However, specifying the correct entity type enables optimized compression for your data characteristics.

K-nearest neighbors search bounded by a time range. Find the K most similar vectors to a query vector, but only within a specific time window. This is a native operation -- not a filter-then-search workaround.

Compares the behavioral path of one entity over time to find other entities with similar patterns, using elastic temporal alignment. Useful for finding vehicles that followed similar routes, sensors with similar degradation patterns, etc.

VectorScaleDB compares an entity's current behavioral trajectory against its historical patterns and fleet baselines to compute an anomaly score. Higher scores indicate more unusual behavior.

Yes. Each tenant gets fully isolated data with separate access controls, audit trails, and usage metering.

AES-256-GCM field-level encryption with envelope key management. Supports local file keys, environment variable keys, and pluggable providers for AWS KMS, Azure Key Vault, and HashiCorp Vault.

Yes. SHA-256 chained audit records for all data operations, queryable via the admin API.

Yes. The GDPR delete endpoint removes all data for a specified entity across all segments, collections, and indexes.

VectorScaleDB uses an embedded storage engine with tiered hot/warm/cold architecture. Minimal resource requirements -- 512MB RAM minimum for small deployments, 4GB+ recommended for production.

Yes. VectorScaleDB deploys as a single binary with embedded storage — no external database, runtime, or container required. Just the binary and a data directory.

Yes. The VectorScaleDB Cloud Platform provides multi-tenant deployment with JWT authentication, API key management, plan-based quotas (Free, Pro, and Enterprise tiers), and a web-based management dashboard. Contact sales@vectorscaledb.com for access.

Enterprise support is available via sales@vectorscaledb.com. This includes dedicated SLA, priority issues, and direct engineering support.

Yes. The MCP server provides 30+ tools across 6 categories that AI agents can use to query, ingest, analyze, and manage data in VectorScaleDB. It supports three operating modes (client, standalone, network) and works with any MCP-compatible client, including Claude Code and Claude Desktop.

Yes. VectorScaleDB provides a full GraphQL API with 55+ queries, 9+ mutations, and real-time subscriptions. It is available alongside the REST, gRPC, WebSocket, and MCP interfaces and is always included.

No. Production is the native single binary — no container or runtime required. Docker is a development-only convenience for local integration testing and cross-compiling Linux artifacts. Published container images and Compose profiles (single-node, multi-node cluster, full-stack with monitoring, edge, and cloud platform) become available at launch; until then, run the binary directly.

The same substrate adapts across a continuous spectrum — from microcontroller-class devices to datacenter clusters — auto-tuning compression, indexing, and resource allocation to the hardware it lands on. There are no brittle fixed size tiers to pick from; the binary detects its environment and configures itself. Three operational modes cover how a node participates in the wider network: fully-federated (joins the cluster and exchanges behavioral templates), lightly-coupled (selective sync), and air-gapped (fully isolated, no outbound connectivity).

Three advances: (1) cross-entity behavioral correlation detects when multiple entities exhibit similar patterns and compresses them jointly for an additional 2–5x, (2) improved multi-stage pipeline with smarter write elimination, and (3) non-vector data types (documents, graphs, blobs) now achieve 30–500x via semantic chunk deduplication.

Content-addressed delivery that routes requests to the semantically nearest cached data, not just the geographically nearest edge node. Combined with semantic chunk deduplication, it reduces redundant storage and transfer for similar data objects.

Yes. The unified storage engine handles vectors, documents, graphs, binary objects, and full-text search in a single layer. All data types benefit from the four-frequency data lifecycle (hot/warm/cold/archive) and can be queried together.

A decentralized trust verification system with threshold signing (no single entity controls the network), behavioral trust scoring, anti-impersonation protections, and an arbiter system for dispute resolution. License revocations propagate network-wide and cannot be reversed.

Yes. All software features compile into a single universal binary — no feature flags needed. Hardware acceleration (CUDA, Vulkan, Metal, NPU) is detected at runtime. Every deployment gets the full feature set: enterprise, clustering, protocols (gRPC, GraphQL, P2P), domain modules (IoT, BCI, genomics, inventory, portals), and all advanced capabilities.

Yes. The VFS (Virtual Filesystem) crate provides cross-platform filesystem mounting. On Windows, it uses ProjFS (Projected File System). On Linux and macOS, it uses FUSE. This lets you browse your database contents as if they were files and directories on your local machine.

The coupling matrix is VectorScaleDB's cross-entity behavioral prediction engine. It learns statistical relationships between entity types and uses those couplings to predict behavior across domains. For example, it can detect that financial market volatility predicts network congestion 2-5 minutes later. The coupling matrix powers cascade prediction, cross-domain alerting, and compression enhancement.

VectorScaleDB enforces 10 inviolable safety constraints at the type-system level: no self-propagation, no single-entity capture (3-of-5 threshold signing), no skill weaponization, irrevocable consent, and bounded self-improvement. These constraints are compile-time invariants that cannot be bypassed by any privilege level, including administrator access.

VectorScaleDB is a modular Rust system organized as a layered architecture: a core type and vector-ops foundation, storage and indexing, compression, query, a server/middleware tier, and a multi-tenant cloud platform — surfaced through client and SDK layers (CLI, Python, MCP, and mobile). Hardware acceleration loads as optional modules at runtime, so the same binary runs everywhere and lights up GPU/NPU acceleration only where it is present.