01
keep the disk, drop the machine
prepare dependencies and input files once, then reuse that setup for repeated tasks or agent evaluations. the saved snapshot remains available after you remove the source sandbox.
back to microsandbox
features / snapshots
save files. resume running workloads.
local snapshots save disk state, or disk, memory, and running processes with --full. cloud currently supports disk snapshots from stopped persistent sandboxes and disk restore.
02
disk or full: choose what survives
disk snapshots save files and owned volumes, then boot a new vm on restore. local full snapshots also save memory and running processes, so execution resumes on restore. full snapshots require a running or paused source and are not yet available in managed cloud.
how it works
capture and restore
01
stop the sandbox
for this disk-snapshot example, stop the sandbox after preparing its files. cloud disk snapshots require a stopped persistent source; local disk capture also supports running sources.
02
save it
with the 0.7.3 cli, use msb snap create with --sandbox. for local archives, msb snap export --with-image includes the base image for offline use; export is not yet available on cloud.
03
restore later
msb snap restore creates a fresh microvm from the saved disk. use msb exec to run your task inside it.
see it work
local: create, archive, and run from a snapshot
microsandbox cli 0.7.3bash
msb stop eval-base
msb snap create swebench-base --sandbox eval-base
msb snap export eval-base:swebench-base --output ./swebench-base.msb --with-image
msb snap restore ./swebench-base.msb --name task-0441
msb exec task-0441 -- pytest -q
msb rm --force task-0441common questions
what is a good use for it?
a warm base image for evals or repeated tasks: set up once, snapshot, then start every run from the same disk.