Durability
Nothing drops.
Every event lands in three regions before we ack. 99.9999% durability, measured.
Horizon — data infrastructure
Est. 2021 · zero lost events
Signal
survives.
Horizon moves, stores and replays your data with no gaps. An event is acknowledged only after three regions hold it. If we ack, it exists.
all regions nominalsoc 2 type iiiso 27001
Trusted by teams that cannot afford gaps
- 01Lumen Pay
- 02Northwind
- 03Orbital
- 04Kestrel
- 05Meridian
- 06Fathom
02 — Product
Six guarantees. One platform.
Every event accounted for, in order, from first write to seven years later.
Replay
Rewind anything.
Any stream, any second, last 7 years.
Speed
8 ms p99.
41M events per second per cluster.
Throughput · last 60 s (simulated)41.2M/s
Order
Strict order.
Per key, per partition, always.
Security
Sealed end to end.
Per-tenant keys, SOC 2 Type II, private regions on request.
soc 2 type iiiso 27001byok
03 — In numbers
Measured, not promised.
99.9999%
durability across three regions
41M/s
events per cluster
8ms
p99 write latency
0lost
events since 2021
04 — How it works
How it works
01 / 03
Connect.
Point any source at Horizon. SDKs, Kafka, webhooks, CDC. Ten minutes, no migration.
Capture.
Each event is sealed, ordered and copied across three regions. You get an ack only after it is safe.
Replay.
Query live, or rewind to any moment. Same data, same order, every time.
05 — One stream, three states
Ingest. Store. Replay.
The same event, seen at each stage of its life. Switch the stage to read what Horizon guarantees there.
Stage 01 · ingest
Every write is acknowledged.
Idempotent writes with typed SDKs. The ack returns only after the event is safe.
- ack8 ms p99
- dedupeidempotent keys
- schemachecked at write
06 — Developer experience
Three lines to never lose data.
Typed SDKs. Idempotent writes. One command to replay.
07 — Observability
Tracks performance in real-time
Hover or focus a node to trace its path.
08 — Integrations
Platforms supported
KafkaPostgresSnowflakeBigQueryS3dbtSegmentStripe
RedshiftClickHouseDatadogWebhooksAirflowFivetranMongoPulsar
09 — Compare
Horizon vs. rolling your own.
| Metric | Horizon | DIY pipeline |
|---|---|---|
| Time to production | 10 minutes | 6 to 12 weeks |
| Durability | 99.9999% | Depends on your on-call |
| Replay window | 7 years | Whatever fits in Kafka |
| Schema drift | Blocked at write | Found in the dashboard |
| Team cost | 0 FTE | 2 to 4 FTE |
10 — Proof
Teams that sleep.
We lost nine hours of payments data in 2023. Since Horizon, none. Not one event.
Replay turned a three-day incident into a ten-minute backfill.
We deleted two services and a pager rotation. The data got better.
11 — Pricing
Pay for signal. Not for noise.
Spark
$0/mo
free forever
- 10M events a month
- 7 day replay
- Community support
Orbit · most chosen
$480/mo
billed monthly
- 5B events a month
- 1 year replay
- 3 region durability
- 24/7 support
Event horizon
Custom
volume pricing
- Unlimited events
- 7 year replay
- Private regions
- 99.99% SLA
12 — FAQ
Questions, answered.
01What does "never loses signal" mean?
We ack a write only after three regions hold it. If we ack, it exists.
02How long does setup take?
About ten minutes. Add an SDK or point a Kafka topic at Horizon. No migration.
03Can I replay old data?
Yes. Rewind any stream to any second within your plan window, up to seven years.
04Where is my data stored?
In the regions you pick. Private regions are available on the top plan.
05What happens if I exceed my plan?
Nothing drops. We bill overage at a flat rate and tell you before it happens.
13 — Start
Questions,
Answers
Free to start. Live in ten minutes. Leave your email and we send an API key and a starter repo.