From first draft
to final approval.

Keep source files, tools, and intermediate results ready for the next round of agent work. Release compute throughout the wait.

Evaluate your workloadExplore the workflows
Research briefsources.jsonAnalysis in progressFindings ready for reviewCode changetest-results.txtBuild and testChanges ready to checkApproval reviewproposed-changes.jsonWaiting for approvalWork ready for a decision

An agent prepares findings, saves its working files, and returns to a review.

Illustrative workload

The follow-up question
has a head start.

Example research task
Compare supplier proposals

Lead times

Payment terms

Missing information

Draft ready for review
Your application produces the analysis. VMYard retains its workspace.

A research report prompts a follow-up: compare another period, examine an exception, or explain the calculation. Keep the working files so the next pass can build on the analysis already done.

  1. Prepare the first result.

    Process approved inputs in a session and return a draft. Keep the scripts and intermediate files behind it.

  2. Save during review.

    Snapshot the workspace while a person checks the result. Release its CPU and memory during the wait.

  3. Answer with the same working context.

    Restore, add the new inputs, and revise. Export the deliverable when it is ready.

Your application provides document processing, source access, citations, and the reasoning behind the answer.

Keep the investigation
with the code.

A failed build leaves clues: logs, generated files, a partial change. A coding agent can use its own Linux workspace to investigate, test a revision, and hand the result back to your development workflow.

Pick up after review.

Save the session while a reviewer examines the change. On return, the source tree, test artifacts, and working state are there for another iteration.

Use the project's actual tools.

Supply an image with the required toolchain and dependencies. Commands can return output or run longer tasks in the background. Your workflow keeps control of review and publication.

What your coding workflow supplies

Repository access, dependency preparation, code review, and the decision to publish remain with your application. A separate VM provides an execution boundary; the permissions and risks of untrusted code still need review for your workload.

Free compute
while approval waits.

Snapshot completed first passes while they wait for approval. Files and guest state stay in storage while CPU and memory serve other work.

After approval, your application authorizes the next action and requests a wake. The session resumes when compatible capacity is available. Business permissions and approval records stay in the system that owns them.

When this pattern is a good fit

This pattern suits meaningful waits and useful intermediate state. Saving and restoring take time and storage. Cross-host moves and recovery copies require guest networking to be disabled.

See how saved sessions
fit your workflow.

Choose a real task, identify the state worth keeping, and compare its next run with a fresh environment.

Build your evaluation briefExplore the session lifecycle
Put the work on your hosts

Choose the environment around your application.

Plan beyond the next run

Understand saved work and recovery after a host is lost.