github.devotd.
NewAI agents

google/ax v0.3.1 — Google's public agent orchestration runtime

The latest release v0.3.1 of google/ax, Google's public runtime that orchestrates multiple agent jobs through sandboxed isolated execution and declarative specs.

Repositorygoogle/ax ↗

The latest release v0.3.1 of google/ax, Google's public runtime that orchestrates multiple agent jobs through sandboxed isolated execution and declarative specs.

GitHub description: Agent Executor (AX) is Google's open agentic orchestration runtime

stars
★ 14K
Forks
730
Language
Go
License
Apache-2.0
Latest release
v0.3.1
Read original source ↗View on GitHub ↗

google/ax (AX, Agent Executor) is an agent orchestration runtime published by Google. An agent here means software that judges on its own and carries out multiple steps. The latest release is v0.3.1, and the changelog for this release does not appear in any verifiable material.

AX is a runtime that runs agent jobs in sandboxes by declaring jobs, workspaces, and model specs, and orchestrates them at scale. Jobs run in isolated sandboxes with CPU and memory limits, and workspaces can be pre-provisioned with repositories, MCP servers, and skills. Network policies are handled through the spec as well. On Kubernetes it deploys on the assumption of Agent Substrate, and the command line supports apply, observe, attach, stop, and resume. The code is written in Go and is released under the Apache-2.0 license. The repository was created on March 30, 2026, and currently has 14010 stars.

Even so, it is better to hold off on judging how far this project can actually go right now. The README describes a development stage in which the core concepts and specs are still being refined, and states that breaking compatibility changes may occur before a stable release. What Agent Substrate means is not explained, and no performance measurements or comparison targets are given. It is also unknown at what scale the isolated execution and large-scale job orchestration have actually been validated. Of the evaluation scores, only momentum is filled in, at 7.4 points. The only thing the current material confirms is that the concepts and specs are being organized.