Announcements

Milestones for the framework, newest first.

RSS feed

ObzenFlow 0.3.0 is on crates.io

0.3.0 is the biggest release of the framework so far. We started with the durable ingress proposal, which kicked all this work off at the end of May. Between 0.2.5 and 0.3.0 alone we’ve released over 30 improvement proposals. The breadth of the framework is now mostly in place, and the next stretch is stabilization and depth.

ObzenFlow has been in development since June 2025, and the first commit already feels like an eon ago. Systems-level development is meticulous, and this release was especially consequential as it lays out the blueprint for the next six months. We have some huge features planned for 2027.

Significant changes since 0.2.5.

  • A public ObzenFlow CLI for tailing the event journals.
  • Complete support for resuming a crashed workflow without re-executing successful side effects.
  • The completed algebraic effect system for isolating non-deterministic code, idempotency keying, and exactly-once processing.
  • Journal-before-acknowledge ingress. A request is written to disk before it is acknowledged.
  • A new merge mode for processing batch data before accepting live streaming input. This is a major feature for teams that will be processing batch and live data in the same flow.
  • Point-in-time joins for matching catalog data that was valid at a specific wall clock time. Think about foreign exchange rates that are only valid during a narrow window.
  • A completely refreshed inference syntax for integrating with Ollama and OpenAI. A single ai! declaration covers both one-shot inference and map-reduce jobs.
  • Enhanced error journaling to support upcoming releases that allow error journals to be joined into topologies as first-class stages.

I’ve been deliberately keeping this project low-key. Durable execution is a complex undertaking, and I am loath to encourage production use of an incubating framework. To set expectations, ObzenFlow will remain a 0.x release indefinitely, until there is evidence that all of its claims hold in production environments.

If you’d like to use this for any kind of production workload, please don’t hesitate to reach out directly for advice.