High-cardinality telemetry ingestion
Routing massive sensor tracks, machine health streams, and device states into shared operational pictures.
181M pts/s intake · 0% drops · bare-metal UDP and HTTP/3SVEIM
One engine · many realtime paths
Overview
Ingest. State. Fan-out. Same box.
Sveim — a single software engine, written from scratch, that delivers high performance and highly efficient data processing in ingest, in-memory storage and fan-out capabilities.
Most live systems still run three separate stacks for the same job. Sveim collapses that path into one self-hosted engine — dense enough that peak traffic stays on hardware you already control.
Scroll
181Mintake · wire pts/s · 0% loss
10Mconnections · 13.2 GB RSS · zero drops
10.65Mmemory ops/s · 256B pipelined
62.2Mwhole platform events/s · 1.78 cores
Re-measured on bare metal · 2026-07-25. Server-side counters, loss reported, load escalated until plateau, cross-validated with two independent clients (agreement within 3%).
Speaks the wire protocols your clients already use. Point them at a new address — measure — then decide.
Capability surface
Protocol & capability surfaceWire protocols spoken · drop-in targets · proven scale
| Capability | Wire protocols spoken | Drop-in replacement for | Proven scale |
|---|---|---|---|
| Intake | UDP telemetry · high-rate ingest · QUIC / HTTP-3 · WebSocket | Kafka · Amazon Kinesis · NGINX HTTP/3 | 181M pts/s |
| Memory | RESP2 / RESP3 · 101+ commands · AOF + RDB persistence · replication | Redis · DragonflyDB · Valkey | 10.65M ops/s |
| Delivery | WebSocket · Pusher protocol · Reverb · HMAC auth · private + presence channels | Pusher · Ably · PubNub · Centrifugo | 10,000,000 connections |
| Spatial | WebTransport / QUIC · raw UDP · authoritative 60 Hz state | Unreal NetDriver · Photon | 174,000 pose/s |
| Media | WebRTC · DTLS · SRTP | LiveKit · Janus · mediasoup | sub-10 µs forwarding |
| Insight | PostgreSQL wire · Arrow FlightSQL · RBAC + TLS | ClickHouse / Druid serving tier | SQL on the live stream |
Benchmarks
Performance — head to headEach Sveim capability against the incumbent product it displaces
| Component | Competitor | Capability | Competitor | Sveim | Advantage |
|---|---|---|---|---|---|
| Key-Value GET (256B) | Redis 7.4 · Valkey 9.1 · Dragonfly | Memory | 3.55M · 4.00M · 0.86M ops/s | 10,652,464 ops/s | 2.7× vs fastest |
| Key-Value SET (256B) | Redis 7.4 · Valkey 9.1 · Dragonfly | Memory | 3.20M · 3.20M · 2.28M ops/s | 7,992,008 ops/s | 2.5× vs fastest |
| Pub/Sub fan-out | Centrifugo 6.8.4 | Delivery | ~130,000 msg/s | 4,600,000 msg/s | 32× |
| Inter-capability transfer | Unix domain sockets | Internal fabric | ~1 µs · syscall + copy | 78M frames/s · 13 ns | ~77× |
| 60 Hz spatial engine | Unreal Engine NetDriver | Spatial | <20 Hz past 128 players | 174,000 pose/s @ 60.0 Hz | 23.4× |
| Edge ingress | NGINX HTTP/3 | Intake | 500 QUIC sessions/core | 3,000 QUIC sessions/core | 6× |
| Metrics ingest, high cardinality | VictoriaMetrics | Intake | 0.5 → 5.6 cores CPU | 0.4–0.5 cores, flat | 2–12× less CPU |
Where it lands
Supported topologies
High-frequency spatial sync
Authoritative state computation and media-adjacent forwarding on a single engine for dense concurrent users.
174,000 pose/s @ 60 Hz · drop-in Unreal NetDriver replacementFlash-spike concurrent fan-out
Synchronizing live odds, scores, and flash inventory during extreme traffic spikes without a separate fan-out fleet.
4,600,000 msg/s · 10,000,000 stable connections · replaces Pusher / CentrifugoMicrosecond order book & tape distribution
Colo-first, ultra-short path routing for financial terminals and book data.
13 ns internal fabric · 78M frames/s · in-process pathHot-state agent & workflow routing
Low-latency event distribution for AI agents and live workflows when standard lakehouses are too slow.
10.65M memory ops/s · drop-in for Redis / ValkeyDense session & alert forwarding
Secure, on-hardware custody for biometrics matching, supervisor views, and live operator screens.
WebRTC · LiveKit · Janus · WebTransport interoperabilitySovereign & air-gapped realtime pipelines
In-country, single-binary deployments for tenants who reject metered, multi-tenant SaaS.
Zero external dependencies · starts in seconds on hardware you controlGetting it running
DeploymentSveim is built to drop into what you already run — not to make you rebuild around it.
Drop-in
Speaks the Redis and Pusher/Reverb protocols your applications already use — existing clients connect by pointing at a new address, not by rewriting code.
No cluster
A single self-contained engine — no JVM, no ZooKeeper, nothing to orchestrate. It starts in seconds on hardware you already have.
Zero downtime
Run it alongside your current stack, mirror real traffic, measure, then cut over when you’re ready. Nothing to rip out on day one.
Hours
From first launch to serving live traffic is an afternoon — a configuration step, not a migration project.
Contact us
Denis Bicer
Founder CTO