Snapshots

Coming (phase 5)

Stop rebuilding the same environment

VM snapshots for CI and agents: prepare a machine once, save it whole, and start every job and agent session from it with services running.

Make every environment start warm

Illustrative. The jobs, commands and timings in it are examples.

01 The snapshot

Do the slow part once. Keep the machine warm. Coming

A clean machine for every job means every job sets up again: the same clone, the same installs, the same warm-up.

So a warm-up job does the slow part once. It clones the repo, installs dependencies, warms the build and starts Postgres.

Then it saves a snapshot of the whole machine, memory and disk both.

zshci.ymlrunner.logsnapshotclaude

job warm-up snapshot.key=app-main

✓ repo cloned

✓ dependencies installed

✓ build warm

● postgres running

▸ saving snapshot memory + disk

✓ snapshot saved app-main

warm-up job, snapshot.key=app-main

repo cloned
dependencies installed
build warm
Postgres running

snapshot saved, memory + disk

02 The fan-out

One snapshot. As many agents as you like. Coming

Each agent session resumes the snapshot app-main, so Postgres is already running when it starts.

Our design target for a resume on Linux is under a second, and any number of sessions can fan out from one snapshot.

It works for Copilot's coding agent, Claude Code, Codex and others. You opt in with a GitHub Actions label option, snapshot.key=….

zshci.ymlrunner.logsnapshotclaude

ohmako machine watch --snapshot app-main

resumes memory + disk, design target under a second on Linux

✓ claude-1 Claude Code postgres up

✓ codex-1 Codex postgres up

✓ copilot-1 Copilot agent postgres up

✓ claude-2 Claude Code postgres up

✓ codex-2 Codex postgres up

✓ agent-1 Your agent postgres up

6 sessions from one snapshot

snapshot app-main6 resumed

snapshotapp-main
memory + disk

repo
deps
build
postgres

Claude Coderesumed
Codexresumed
Copilot agentresumed
Claude Coderesumed
Codexresumed
Your agentresumed

How it works

Prepared once, restored every time.

A warm-up job clones your repositories, installs dependencies and starts your services. Then the whole machine is saved, memory and disk, and later jobs and agent sessions start from it with everything already running.

One snapshot, many machines

Any number of independent machines restore the same snapshot. An agent session gets one over MCP, or with ohmako machine at a terminal.

Fresh when your code changes

A snapshot is invalidated when its inputs change, so nothing starts from a stale environment.

No secrets inside

Secrets are never captured in a snapshot that's reused.

Restore time

Our design target is under a second for a Linux restore. We'll publish measured times once it's benchmarked.

Price

Pay to restore and to keep.

Restoring

Coming

$0.04 per restore

Keeping one

Coming

$0.025 per snapshot-hour

Join the invite list

Tell us which products you want first. Invites go out in waves, product by product.

Make every environment start warm