Google's AX v0.3.0 splits its open agent runtime into three services

The Apache-2.0 runtime moves task state to Redis Streams to survive millions of short-lived agents; Apple's M5 Ultra also outruns the DGX Spark on local LLMs.

Nowline SEP 21 7:00 PM banner

Top AI stories from the last hour

Top AI stories from the last hour

Copy markdown

  • State moves off etcd, onto Redis Streams

    v0.3.0 splits AX into three services — an API frontend, a reconciler, and a sandboxed task runner — and moves task state out of Kubernetes custom resources into Redis Streams, because etcd was never built for the churn of millions of short-lived agent tasks. If you've been bending Kubernetes to run agents, this is the pattern to copy.

  • Apache-2.0, and one command to try it

    AX is a distributed agent runtime with native recovery and resumption, so a long-running or crashed agent picks up where it left off. Install it with `go install github.com/google/ax/cmd/ax@latest`, then run `ax --input "..."`. It topped Hacker News at 481 points.

  • Elsewhere: Apple's M5 Ultra outruns the DGX Spark for local LLMs

    Tom's Hardware clocked the M5 Ultra Mac Studio's prompt processing faster than Nvidia's DGX Spark, and its throughput nearly 4x higher across the board on Qwen 3.8-27B, with 1.2 TB/s of bandwidth and up to 256 GB of unified memory — enough to hold very large models on a single node. For local inference, one Mac is now a serious box.