Infrastructure for AI agents

Agents need
room to run.

Run agents in isolated workspaces.
Save their progress between jobs.

Join the waitlist See how it works

Give each agent its own workspace.

Illustration
Start work. Save progress. Resume later.Take a closer look
Save work between tasksWorkspaces

Keep progress.
Release compute.

Give each agent its own workspace. Save the session before it sleeps, free CPU and memory between jobs, then continue from the saved state.

Session / research-agentRunning
/worknotes.md   sources.json   analysis.py
Agent workspace 
$ python /work/analysis.py

Reading sources.json
Working through 24 documents…
Saved findings to /work/notes.md

Workspace ready for the next step.
ComputeAllocated

Follow live output and stop a task when needed.

Session snapshotState preserved. Compute released.

Interactive example using snapshot sleep. A simple pause keeps memory allocated.

Explore the session lifecycle
Use your capacity wellCapacity

Make room
for more work.

Distribute sessions across your servers. Put released compute back to work and reserve capacity for saved sessions to resume.

Move saved work. Move eligible saved sessions to a compatible server when the original needs maintenance. See how recovery works.

Keep a recovery copy. Keep a copy of an eligible saved session on another server. Recover from that copy after a server fails.

Plan for work to resume. Reserve CPU and memory for returning sessions as you place new work.

Plan your deployment
Control each workloadControl

Keep each
agent separate.

Run agent tools and code in separate environments. Set limits on what each session can consume and keep track of its results.

A workspace for every session

Each session runs in its own Linux microVM, with separate files and processes for the agent's work.

Control resource use

Allocate CPU and memory, set task timeouts, and cancel work that needs to stop. Apply quotas when teams share infrastructure.

Avoid blind retries

If a command's result is unknown, VMYard reports it instead of automatically running the command again.

Explore workload controls
From ERP.AI Labs

Give your agents
room to run.

Run code in isolated workspaces. Keep state between jobs. Resume when there's work to do.

Join the waitlist

Plan your evaluation