Streamcoil
Managed Apache Kafka, run by the people who page for it

Streamcoil is a fully managed Kafka backbone — multi-zone clusters, exactly-once delivery, and a control plane that rebalances brokers before you feel the lag. You wire up producers and consumers. We carry the pager, the partitions, and the 3 a.m. failover.

  • 100% open Kafka protocol
  • 99.99% uptime SLA
  • Multi-zone, every cluster
producer.py — your code, unchanged
from confluent_kafka import Producer

p = Producer({
  "bootstrap.servers": "prod-orders-eu.streamcoil.io:9092",
  "enable.idempotence": True,   # exactly-once, on by default
})

p.produce("orders", key=order_id, value=payload)
p.flush()   # → committed to 3 zones before it returns

Carrying the order pipelines, ledgers, and real-time analytics that can't stop

Northwind RetailCadence LogisticsVerge MobilityHalcyon BankLumen MediaAtlas TelemetryNorthwind RetailCadence LogisticsVerge MobilityHalcyon BankLumen MediaAtlas Telemetry
The backbone

One streaming spinefor every event yoursystems emit.

Kafka shouldn't be a second on-call rotation. Streamcoil owns the brokers, the rebalances, the version upgrades, and the storage tiering — your team ships pipelines and leaves the partitions to us.

Multi-zone, self-healing brokers

Every cluster spans three availability zones with rack-aware replicas. When a broker dies, the control plane reassigns its partitions and re-replicates in the background — leadership moves in seconds and producers see a blip, not an outage. No runbook, no manual partition surgery, no pager at 3 a.m.

Exactly-once, end to end

Idempotent producers and transactional reads ship on by default. A payment event lands once, is processed once, and never replays into a double charge.

Tiered storage, infinite retention

Hot data stays on fast local disk; cold segments tier to object storage automatically. Replay six months of events at object-store prices, not broker-disk prices.

Schema registry, enforced

Avro, Protobuf, and JSON Schema with compatibility checks at produce time. A breaking change is rejected before it ships — not discovered by a consumer at midnight.

Elastic partitions

Take a topic from six partitions to six hundred with one call. The platform reshuffles load in the background and keeps per-key ordering intact the whole way.

Measured at the broker, not the brochure

18M+
Messages/sec per cluster
4 ms
Median produce latency
99.99%
Uptime SLA
3 AZ
Replication, every cluster
Wire it into everything

Connect the wholestack without writingglue code.

150+ managed connectors move data in and out of your databases, warehouses, and SaaS tools — provisioned, monitored, and dead-letter-handled by Streamcoil, not by a cron job nobody remembers owning.

Source & sink connectors

Stream change-data-capture out of Postgres, MySQL, and MongoDB; land it in Snowflake, BigQuery, S3, or Elasticsearch on fully managed Kafka Connect.

Stream processing, no cluster to babysit

Author ksqlDB and Flink jobs straight against your topics. We run, scale, and checkpoint the stateful workers; you write the query and watch the results land.

Dead-letter queues that explain themselves

Failed records route to a DLQ carrying the full error, payload, and offset — readable in the console and replayable the moment you ship the fix.

Change one URL, not your code

100% Kafka wire protocol. Point librdkafka, Spring, Go, or your current client at a new bootstrap address and every service connects unchanged.

What teams build on the current

The same log carries the sale, the check, and the dashboard.

One ordered, replayable stream feeds the order that just got placed, the fraud check that clears it, and the screen watching it all move.

Commerce

Order & payment pipelines

Every checkout, refund, and inventory move flows through one ordered, exactly-once log — so fulfilment, billing, and analytics all read the same truth in the same instant.

Data platform

Change-data-capture fan-out

Mirror operational databases into the warehouse, the search index, and the cache from a single CDC stream — instead of six brittle batch jobs that drift apart overnight.

Risk

Fraud & anomaly scoring

Score transactions in flight with stream processing, holding model state in the log so a replay rebuilds the exact decision after every deploy.

Platform ops

Telemetry & observability bus

Funnel metrics, logs, and clickstream into one high-throughput pipe, then tee it to storage, alerting, and live dashboards without re-instrumenting a single service.

From the on-call channel

Teams stopped owning Kafka. They still own their data.

We ran our own three-broker cluster for two years and spent every quarter firefighting rebalances. We cut over to Streamcoil in a weekend and the platform team got their on-call rotation back. Throughput doubled and the pages stopped.

D
Dana Okafor
Staff Engineer, Northwind Retail

Exactly-once stopped being a slide and started being real. We push millions of payment events a day and haven't reconciled a single duplicate since the migration. That line item alone paid for the contract.

M
Marcus Reid
Head of Platform, Halcyon Bank

Tiered storage let us keep a full year of events for replay at a price our old cluster couldn't touch. When we rebuilt a downstream service, we replayed history from offset zero and it just caught up.

P
Priya Anand
Principal Data Engineer, Atlas Telemetry
Pricing

Pay for throughput, not for seats.

Billed on the data you actually stream and store. No per-broker tax, no per-developer license, no charge for the connectors you switch on.

Developer

For prototypes and the first pipeline.

$0/mo
  • Single-zone cluster
  • Up to 50 GB/mo throughput
  • Schema registry + 5 connectors
  • 7-day retention
  • Community Slack support
Most popular

Scale

For production traffic you can't afford to drop.

$0.11/GB streamed
  • Multi-zone, self-healing clusters
  • Exactly-once + transactions
  • Tiered storage, infinite retention
  • 150+ managed connectors
  • 99.99% uptime SLA

Enterprise

For regulated, multi-region streaming fleets.

Custom
  • Dedicated VPC or BYOC deployment
  • Multi-region cluster linking
  • Private networking + SSO + audit logs
  • SOC 2 Type II + HIPAA
  • Named streaming architect on call

Engineer's questions, answered.

Is this real Apache Kafka or a look-alike?

It's 100% the open Kafka wire protocol. Your existing librdkafka, Java, Spring, Go, or Python clients connect by changing one bootstrap URL — no SDK swap, no rewrite. We manage the brokers, the ZooKeeper-less KRaft control plane, storage, and upgrades; the API your code talks to is plain Kafka.

What does exactly-once actually guarantee?

Idempotent producers and transactional consumers are enabled by default, so a record is written once and read once even across producer retries and consumer restarts. For a payment or order event, that means no duplicate processing and no lost message inside the streaming layer — the property holds end to end across a transaction.

How does a broker failure affect my producers?

Every partition is replicated across three availability zones. When a broker fails, the control plane elects a new leader from an in-sync replica in seconds and re-replicates the lost copies in the background. Clients following the protocol retry to the new leader on their own, so the worst case is a brief latency blip — not data loss and not an outage.

Can I keep months of data for replay without going broke?

Yes. Recent segments stay on fast local disk for low-latency reads, while older segments tier automatically to object storage. Set infinite retention and replay from any offset at object-store prices, which turns rebuilding a downstream service or backfilling a new consumer into a routine task instead of a risky one.

How do I migrate off my self-managed cluster?

Our mirroring tool replicates topics, consumer-group offsets, and schemas from your existing cluster with no downtime. You cut producers over topic by topic, watch lag stay at zero on the live dashboard, then point consumers at the new bootstrap URL. Most teams finish a full migration in a single weekend.

Send your first event in five minutes.

Spin up a multi-zone cluster from the console, or talk to a streaming architect about moving your production traffic. No sales call required to start building.