Apache Kafka compatible · object-store native

Topics are free.Partitions are free.

Every other broker bills you for the shape of your cluster — per partition, per broker-hour, per cluster-hour. Kodansu bills for data. Create a hundred thousand topics; the invoice doesn't notice.

Request a cluster Read the source

Speaks the Kafka wire protocol · Stores on S3 or GCS · $0 while nothing is connected

The pricing question

Everyone charges for the cluster. We charge for the data.

This isn't a discount — it's a different unit. A broker that keeps no state on disk has no reason to price the topology, so it doesn't.

Broker	What the meter runs on	What a new topic costs you
Confluent Cloud	partitions · cluster-hours · GB in/out · storage	more partitions, more meter
Amazon MSK	broker-hours · provisioned storage	capacity you sized in advance
Redpanda Cloud	cluster-hours · partitions · GB	more partitions, more meter
Kodansu	bytes written · bytes read	nothing

A topic per tenant. A topic per device. A topic per agent session. Shapes that are absurd on a partition-priced broker are just how you'd model it, if modelling it that way were free.

Why we built the storage layer the way we did

How it holds up

Three decisions, and the price follows from them.

Brokers keep nothing

All state lives in the object store. Any replica answers any request, so there is no rebalance, no partition ownership, and nothing to warm up when one starts.

stateless

Topics share segments

Many topics coalesce into one object write instead of each paying for its own. Object-store requests — the thing that actually costs money — stop scaling with how many topics you have.

coalesced segments

Idle really is idle

No disks to keep warm, and a load signal that reads a fleet holding a thousand idle long polls as doing no work. Nothing connected, nothing running, nothing billed.

scale to zero

Straight answers

What's real today, and what isn't.

Kodansu is at 1.0.0-alpha. It runs our own change-data-capture production workload — thousands of topics, continuously. It has not yet run yours. Here is the honest line between the two.

Shipped and in use

✓Kafka wire protocol, generated from the official message descriptors
✓Java, librdkafka and Go clients exercised in CI against several Kafka versions
✓Consumer groups, compaction and retention
✓SASL/SCRAM and TLS
✓ACLs enforced on produce, fetch and the admin APIs
✓Client quotas over the standard admin API — so one tenant can't spend another's budget
✓S3 and Google Cloud Storage

Not yet — ask us before you count on it

—An SLA. We're hand-running these clusters, and we'll say so rather than sell you a number.
—Azure Blob Storage
—Self-serve signup and billing
—Published pricing — we're measuring our own cost first, with the first teams on it
—Multi-region clusters

Concierge access

We provision itby hand.

No signup form that drops you into an empty console. Tell us what you want to run, and we set the cluster up ourselves — bootstrap URL and credentials back to you within one business day.

We're taking about ten teams to start, because we intend to actually talk to each one. Bring us something that breaks it.

Free while we're in concierge.
No card, no contract, and no obligation to keep using it.

Work email *

What you'd run on it *

Preferred region

Roughly how many topics

Client library

Anything that would break it

Opens your mail client with the answers filled in — nothing is submitted anywhere else.