Jitsu

An open-source event data platform — collect events from sites, apps and servers, then land them in a warehouse and in third-party tools. Positioned explicitly as a Segment alternative. MIT, 4.9k stars, 367 forks; Go and TypeScript.

This is the spoke’s first source on the layer above the engine: not how an analytical database stores or queries data, but how the data gets into one.

Architecture

Five independently scalable services, which is worth listing because the split names the problems:

Ingest (Go) — the HTTP endpoint that accepts events.
Rotor (TypeScript) — routing and user-supplied function execution.
Bulker (Go) — warehouse ingestion, batching, and schema management.
Sync-controller (Go) — orchestrates pulls from third-party sources, using Airbyte-compatible connectors.
Console (Next.js) — admin UI and APIs.

It runs on Postgres, Kafka or Redpanda, clickhouse, and MongoDB. Self-hosting wants Kubernetes.

Destinations

clickhouse, BigQuery, snowflake, Redshift, Postgres, S3, GCS, and assorted SaaS tools.

Note what that list is: the spoke’s own subject matter, addressed as interchangeable endpoints. From the pipe’s point of view clickhouse and snowflake differ in their loading interface and not much else — a flattening the spoke’s storage-compute-disaggregation axis says should matter and, at this layer, apparently does not.

The claim: latency, and where it comes from

Jitsu’s pitch against Segment is delivery speed — events landing “in minutes, not hours,” micro-batched or row-by-row depending on destination, against Segment’s once- or twice-daily warehouse loads.

That is a claim about the loader, not the engine, and it bears on this spoke’s open question about how much of the gap is the engine versus the managed service (tinybird raised it as a serving problem). Jitsu is the same question at the other end of the pipeline: ingestion latency is set by the pipe’s batching policy, and a fast columnar engine behind a twice-daily loader is a slow analytics stack. No measurements are given — “minutes, not hours” is a positioning statement, and the comparison is to a competitor’s documented behaviour rather than a benchmark either party ran.

Pricing as an architectural argument

Segment bills per Monthly Tracked User; Jitsu bills per event volume in cloud, and self-hosted has no usage limit at all under MIT with no feature gating. The free cloud tier is 200k events/month.

Recorded because the spoke’s other vendor source (tinybird) makes its case on operational burden, and this one makes a structurally similar case on billing shape. Both are arguments that the commercial packaging around an engine is the real product.

Tier: T3

The project’s own repository and README, from a company selling the hosted version. Usable for what the system does and how it is built; the Segment comparison is competitive positioning by an interested party and nothing here is independent.

clickhouse · snowflake · tinybird · analytical-databases · storage-compute-disaggregation