AX

Declarative runtime for running agentic tasks at scale with sandboxing and state management

Automation & Agents Agent OrchestrationContact for Pricing

40KMonthly visits

25Popularity

About AX

AX provides a declarative runtime for executing agentic tasks at scale with strict isolation and resource control. It introduces four core primitives—Task, Workspace, Gateway, and Model—that handle execution, environment setup, network policies, and configuration in a unified way. Tasks run in lightweight sandboxes with CPU and memory limits, allowing for cheap creation, suspension, and disposal. Workspaces automate the setup of Git repositories, MCP servers, and dependencies before a task begins, while Gateways enforce network policies through allowlists and credential injection. The Model primitive centralizes configuration for models, parameters, and secrets, enabling easy rotation or version pinning. AX runs on Agent Substrate, a compute runtime designed for high-density stateful actor lifecycles, supporting billions of concurrent tasks with sub-second resumption and dense multiplexing of idle agents. The platform integrates generative AI features, such as setting up workspaces via plain English descriptions, and supports interactive coding agents, long-running servers, Jupyter notebooks, and headless browser testing.

Key features

Task execution with sandbox isolation
Workspace setup for Git repos and dependencies
Network policy enforcement via Gateway
Model configuration and secret management
Sub-second agent resumption
Dense multiplexing of idle tasks
Generative workspace setup
Interactive coding and headless browser testing

Use cases

Running reinforcement learning loops at scale
Evaluating agentic systems with reproducible sandboxes
Setting up isolated development environments via natural language

Pros

Declarative configuration using YAML for tasks, workspaces, and policies
Lightweight sandboxing with CPU and memory limits for untrusted code
Sub-second suspension and resumption of idle agents
Generative workspace setup via plain English descriptions
Dense multiplexing of tasks to optimize resource usage

Cons

No explicit mention of a free tier or open-source availability beyond GitHub
Limited to declarative YAML configuration without visual interface
Requires Agent Substrate as underlying runtime
Focused on agentic workloads, not general-purpose orchestration

Frequently asked questions about AX

What is AX and what does it do?

AX is a declarative runtime for executing agentic tasks at scale with strict isolation and resource control. It provides primitives like Task, Workspace, Gateway, and Model to handle execution, environment setup, network policies, and configuration in a unified way.

Who is AX designed for?

AX is designed for developers and researchers building agentic systems, including interactive coding agents, long-running servers, Jupyter notebooks, and headless browser testing. It is also suitable for running massive fleets of agents for tasks like reinforcement learning and agent evaluation.

How does AX handle workspace setup?

AX automates workspace setup by allowing users to either list Git repositories, MCP servers, and dependencies or describe the goal in plain English. AX then prepares the environment automatically before the task starts.

What are the core primitives in AX?

The four core primitives in AX are Task for isolated execution, Workspace for environment setup, Gateway for network policies, and Model for centralized configuration of models, parameters, and secrets.

How does AX manage network policies?

AX enforces network policies through Gateways, which allow users to define and manage allowlists of hosts and ports, as well as inject credentials to incoming requests.

How do I get started with AX?

To get started with AX, users can declare an agentic task using a YAML file, apply it using the AX CLI, and then monitor or interact with the task. Detailed examples and documentation are available on the AX GitHub repository.