AX AX
AX (Agent Executor) is Google's open-source distributed harness runtime that dynamically provisions isolated, suspendable/resumable environments to execute agents and harnesses at scale across a cluster. It provides a Kubernetes-like, declarative orchestration layer for running large numbers of autonomous agent workloads.
Repository created 2026-03-30 and currently in active early development (~1,967 stars, 116 forks, 35 open issues); the team has temporarily paused external pull requests while stabilizing core architecture, resumption protocols and runtime specifications, and encourages issues for feedback.
As of
- Distributed runtime with isolated execution for harnesses, skills, tools and agents
- Automatic recovery and execution resumption from failures or interruptions
- Single-writer architecture with a durable event log for consistent state management
- Built-in support for frontier harnesses plus custom implementations
- Portability across small and large deployments, scaling to billions of tasks per cluster
- Customizable environments, MCP tools, skills and instructions
- Runs on top of Agent Substrate for sandboxed execution
Platform and infrastructure engineers who need to orchestrate large-scale, long-running autonomous agent workloads with reliability and resumability, especially teams already comfortable with Kubernetes-style declarative systems.
Developers wanting a stable, production-ready API today, or those without the infrastructure and model-provider budget to run distributed agent clusters.
AX is an open-source project released under the Apache License 2.0 and is free to use; costs come from the underlying compute, sandboxing (Agent Substrate) and model providers you connect. It is in active early development with breaking changes expected before a stable release.
AX offers a compelling open-source foundation for running massive numbers of autonomous agents with strong resumption and consistency guarantees, backed by Google and an Apache 2.0 license. However, it is explicitly pre-stable with major breaking changes expected and external contributions paused, so it is best suited to experimentation and infrastructure-forward teams rather than production deployments today.
Is AX free?
Yes - AX is Open Source. AX is an open-source project released under the Apache License 2.0 and is free to use; costs come from the underlying compute, sandboxing (Agent Substrate) and model providers you connect. It is in active early development with breaking changes expected before a stable release.
Is AX open source?
Yes - AX is open source. AX is an open-source project released under the Apache License 2.0 and is free to use; costs come from the underlying compute, sandboxing (Agent Substrate) and model providers you connect. It is in active early development with breaking changes expected before a stable release.
Who is AX best for?
Platform and infrastructure engineers who need to orchestrate large-scale, long-running autonomous agent workloads with reliability and resumability, especially teams already comfortable with Kubernetes-style declarative systems.
Who is AX not ideal for?
Developers wanting a stable, production-ready API today, or those without the infrastructure and model-provider budget to run distributed agent clusters.
What are the best AX alternatives?
The closest AX alternatives on ai.dosa.dev are Jules, OpenHands, SWE-agent - all listed under Autonomous & Async Agents.
Who makes AX?
AX is developed by Google. It is listed in the Autonomous & Async Agents category on ai.dosa.dev.
Favorite this tool to revisit it later, or Zap it to contribute to the public vote count.
Content on this page is AI-generated. Please verify details with the vendor's website for accuracy.