Vectorio Protor — the data plane we wished existed.

02b · code or canvas

Write it in VXL.
Or wire it on the canvas.

VXL is a terse, typed language for workflows — every node, property and edge as plain text. Prefer to see it? The same workflow opens on the canvas and runs live, rows moving through each node.

VXL .viow JSON one workflow · two open formats

climate-music-continuous.vxl// Climate Music — Continuous Sonification
pipe "climate-music-continuous" v1 {

  // LAYER 0 · sequence generator — 8 weekly date windows
  node "date-sequence": gen.sequence {
    .mode      = "DateRange"
    .autoStart
    .startDate = "2025-01-01"
    .endDate   = "2025-03-01"
    .dateStep  = "7.00:00:00"
  }

  // LAYER 1 · build the API endpoint from each window
  node "derive-endpoint": etl.compute {
    .expr = CONCAT("/v1/archive?lat=40.71…&start_date=",
                 col"start_date", "&end_date=", col"end_date")
           as "endpoint_url": String
  }

  // LAYER 2 · advance one window per completed fetch
  node "endpoint-modulator": dyn.property {
    .triggerMode        = "OnFact"
    .watchSourceNodeIds = ["audio-output"]
    .watchFactTypes     = [2024]
  }
}

climate-music-continuous.viow// the same workflow, serialised — standards-friendly JSON
{
  "$schemaVersion": "1.0",
  "Id": "climate-music-continuous-001",
  "SyncFlowSettings": { "StreamingMode": "True", "ErrorHandling": "Continue" },
  "Elements": [
    {
      "TypeKey": "Vectorio.Generators.Sequence",
      "Id": "date-sequence",
      "Configuration": {
        "Mode": "DateRange", "AutoStart": "True",
        "StartDate": "2025-01-01", "EndDate": "2025-03-01",
        "DateStep": "7.00:00:00"
      }
    },
    {
      "TypeKey": "Vectorio.ETL.Compute",
      "Id": "derive-endpoint",
      "Configuration": { "StreamingMode": "True",
        "Expressions": { "endpoint_url": "CONCAT('/v1/archive?lat=40.71…')" } }
    }
  ]
}

The Vectorio workflow designer — nodes wired on a canvas, running live at 40 rows per second, edges animating between Sequence, Compute and Publisher nodes — Workflow designer · running · 40 rows/s

Same workflow, same signed .viow — author it as text or on the canvas, run it and watch the rows move.

scan · single-thread

74.79GB/s

ForBitpack columnar decode

group-by · multi-agg

112GB/s

fused SUM + MIN + MAX + COUNT · CPU, no GPU

cross-machine

22GB/s

VSPP · gRPC + TCP plane

seal-time agg

15B rows/s

single-aggregate path · O(N)

01 · architecture

One plane,
three concerns.

Movement, compute, and distribution are designed as one substrate. They share batches, columns, kernels, and signals. There is no glue, no bridge, no shim — by intent.

Sources

Postgres→ Kafka→ REST→ CSV / Parquet→ FastTable mount

Movement

streaming runtime· backpressure· alive-budget· VSPP transport

Compute

SQL · DML · UDF· columnar kernels· radix group-by· bitmap filter· radix join

Distribution

cluster nodes· control plane· zone-map pruning· mTLS + AES-256

Sinks

Lakehouse→ REST→ Forensic pack→ App / Studio

Same diagram, scale-free: from a single laptop pipeline to a 10,000-node cluster, the plane composition does not change — and the same plane carries 200+ nodes. See the catalog ↓

01b · the catalog

200+ nodes.
One open format.

Over 100 connector families ship with the engine — databases and warehouses, every major cloud, streaming, files, SaaS, AI/ML (local too), DSP & media, geospatial, healthcare, security & compliance, observability. Every node is a signed extension; every workflow is open .viow text. No node leaves the box.

Databases & warehouses

PostgresMySQLSQL ServerSQLiteMongoDBRedisClickHouseSnowflakeBigQueryElasticsearch

Cloud & object storage

S3Azure BlobAWS SQS/SNSDynamoDBLambdaGoogle Drive/SheetsSFTP

Streaming & messaging

KafkaAMQPRabbitMQMQTTRESTGraphQLODataWebServer

Files & formats

CSVJSONXMLExcelParquetBinaryFileDocumentsText

SaaS & business apps

SalesforceHubSpotShopifyStripeServiceNowJiraConfluenceLinearSlackTelegram

AI & ML

Local LLMOllamaOpenAIAI VisionStable DiffusionClassificationChain-of-Thought

DSP, audio & media

DSP analysisfrequencyAudio productionVoiceSonifyImage processingVideo

Geospatial

Spatial analyticsGDALTilesSTACSentinel

Healthcare & identity

FHIRHL7Pathology (ONNX)Active DirectoryLDAP

Security, compliance & governance

PolicyGate · ABACNIS2Forensic packLineageQualityGateCheckpoint

Observability & ops

DatadogSplunkSentryPagerDutyCloudWatch

Orchestration & control

Execution controlProperty controlCommand controlFact-opsSub-workflows

A representative slice of 100+ connector families / 200+ nodes. Don't see yours? The connector API is ~50 lines — a new node is a day, not a sprint.

02 · workflows are text

Diff it.
Sign it. Ship it.

A Vectorio workflow is plain text — authored in VXL, serialised to a standards-friendly .viow JSON. It lives in your repo, your code review, your audit trail. No vendor blob, no proprietary editor required.

Git-native. Branches and pull requests, not "exports".
Human-readable. Reviewers see what changed without opening a tool.
Open formats. VXL for authoring, JSON for everything else — no lock-in.
Signable. The same signature primitive that protects extensions protects pipelines.
Composable. Workflows reference workflows. Reuse without abstraction taxes.

03 · benchmarks

Numbers an auditor will trust.

Benchmarks below are reproducible from the open harness. We publish hardware, kernel mode, branch, and commit. We do not pre-aggregate seed data to flatter a chart.

Workload	Kernel	Hardware	Throughput	vs. last release
Columnar scan	ForBitpack decode · 1 thread	M-class, 16 cores	74.79 GB/s	+12.4 %↑
GroupBy (single agg)	radix · adaptive bank	M-class, 16 cores	15.0 B rows/s	+9.6 %↑
GroupBy (multi-agg)	fused SUM/COUNT/MIN/MAX	M-class, 16 cores	7.7 B rows/s	+18.1 %↑
GroupBy (fused, throughput)	byte-key dense · SUM/MIN/MAX/COUNT	M-class, 16 cores	112 GB/s	new★
Distributed query	VSPP · gRPC + TCP	10× workers	22.0 GB/s	+38 %↑
Distributed query (D11b)	VSPP raw bypass	10× workers	815 M rows/s	new★
Mounted store vs CSV	FastTable mount	M-class, 16 cores	298 ×	+212 %↑
AES-256 GCM transit	signed VSPP	M-class, 16 cores	1.8 GB/s	stable

All measured on single hardware — scale to a cluster without re-architecting, no managed service, no per-query bill. Reproducible from our public benchmark harness.

04 · subsystems

A field guide to the engine.

SUB · 01

Streaming runtime

Backpressure-aware, alive-budget controlled, NUMA-aware page allocator. Per-source byte budgets, end-to-end memory control.

SegmentPool·AliveBudget·VSPP

SUB · 02

SQL compute

Columnar kernels, fused multi-aggregate, bitmap filter, radix join. SQL DML, distributed pruning over gRPC.

RollupNode·QualityGate·ComputeNode

SUB · 03

Distributed plane

Cluster.Node + ControlPlane + VSPP data channel. mTLS + AES-256, zone-map pruning, JWT/ABAC, tenant isolation.

ControlPlane·WorkerStore·165 tests

SUB · 04

Plugin architecture

System extension hooks, signed extensions, contract validation, command contribution. Probes, actuators, contracts.

AssemblySignatureVerifier·SignaturePolicy

SUB · 05

Forensic layer

Every query leaves an auditable record — evidence collector, hash chains, signed manifests, replayable lanes — so you can answer "why was this number calculated?" to any auditor.

Vectorio.Forensic·EvidencePack

SUB · 06

NodeDataStore

Unified persistence layer. Store, dedup, spill, update policy, load strategy. Distributed fields reserved in manifest.

FastTable / Store·UpdatePolicy

SUB · 07

Studio & Designer

VSCode-style docking, panel catalog, theme integration, layout persistence. Visual composer over the same engine.

DockFramework·SP3

SUB · 08

App Builder

Compose .vapp, ship to iOS / macOS / desktop. Fact-driven binding, event pipeline, design system, secure deployment modes.

AppBuilder·VecNodeControl

Any sector. One engine.

DEF · 01 Defense & aerospace air-gapped · SCIF-capable · sealed transfer

FIN · 02 Banking & capital markets MiFID II · Basel III · DORA

BIO · 03 Clinical & biotech HIPAA · GxP · 21 CFR 11

GOV · 04 Public sector & civic NIS 2 · Schrems II · sovereign

ENE · 05 Energy & grid IEC 62443 · KRITIS · SCADA-class

SAT · 06 Earth observation raster tiles · spatial primitives · signed

IND · 07 Industrial & manufacturing edge · existing hardware · deterministic

RET · 08 Retail & commerce PCI-DSS · peak-day · one host

Sovereign by design.

Engineered in Berlin. Runs on the hardware you own, in the jurisdiction you choose. Every extension that touches the pipeline carries cryptographic provenance — no managed lock-in, no vendor cloud required.

On-prem, on-device, or sovereign cloud. The same engine, the same kernels. No managed lock-in. DEPLOYMENT · cross-platform · containerised · bare-metal
Signed extensions, verified at load. Every plugin carries a publisher signature. Policy decides what runs. SECURITY · AssemblySignatureVerifier · trusted publisher PEM
Forensic-grade audit out of the box. Evidence packs, hash chains, signed manifests — built into the data plane, not bolted on. AUDIT · Vectorio.Forensic · evidence collector / pack / verifier
Plain workflows, plain JSON. .viow is a text format. Diffable, reviewable, signable. No vendor blob. FORMAT · open · text · git-native

05 · one of two

Protor is one half of Vectorio.

Protor is Vectorio's data engine — the deterministic plane your pipelines move and compute on. Its independent sibling, Indago, is our discovery engine: the same company, the same sovereignty principles, a different problem entirely.

Two products, one engineering bar. If you came for the data plane you're in the right place — Protor is what this page is about.

Talk to us → Meet Indago →

Numbers you can defend.

Write it in VXL.Or wire it on the canvas.

One plane,three concerns.

200+ nodes.One open format.

Diff it.Sign it. Ship it.