AX and Agent Substrate, together

AI agents tend to work in short bursts, then spend most of their time waiting. Holding a sandbox open for every waiting agent wastes resources. Agent Substrate is built to fix that, and AX runs on top of it.

Agent Substrate runs agents on Kubernetes, and calls each one an actor. Workers are pods, started ahead of time. When work arrives, an actor is resumed on a worker, in its own sandbox. When it's suspended, Substrate saves a snapshot and frees its spot. So a few workers can serve many actors.

AX is the orchestrator. You describe the job in YAML, much like Kubernetes. A Task says what to run. A Workspace says what it needs first: repos, tools, and skills. And a Model sets the AI model AX itself uses.

Here's how they work together. You apply a Task, and AX registers it on Substrate as a suspended actor. Resume it, and Substrate runs it on a worker. Inside the sandbox, AX's runner prepares the workspace first, then starts your command.

Agent gone quiet? Suspend it. AX has Substrate save just the workspace folder, and the worker spot is freed. Resume later, and the same files come back in a fresh container. Files survive, but running programs start over. So save progress to the workspace.

AX says what to run. Substrate decides where it runs. And your agent does the actual work.