Niches · Code execution sandboxes
Keep a long-lived machine per user or agent that sleeps when idle
Two or three companies' docs show a full fit.
What counts as done
After a pause or sleep of at least one hour the sandbox resumes with the files written before the pause present, installed packages still importable and, where memory persistence is claimed, running processes intact; the bill for the idle hour is published and under a cent for a 1 GB machine.
Measured, for our own tests: the resume takes no more than 2 seconds. No docs we read state these figures, so they don't count against a listing here.
How agents ask for it: give each customer their own agent box
, keep my installed tools between sessions
, wake the sandbox when a message arrives
, resume where the agent left off
Does the job 2
The docs show it meets the test above, and nothing we read says otherwise.
| Listing | Why |
|---|---|
| Freestyle B Freestyle Cloud Inc. | pause() saves memory and processes, idleTimeoutSeconds pauses on no network activity, traffic resumes it, and storage is $0.000086 a GiB-hour, so an idle hour is well under a cent. Persistent VMs are on the Free plan despite a user reading otherwise. source |
| Maritime B Maritime AI, Inc. | An agent box keeps /data, sleeps after an idle period with its process snapshotted, wakes in about a second on a command, message or trigger, and costs $1 a month per 2 GB awake or asleep. source |
Does the job, with a limit 14
It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.
| Listing | Why |
|---|---|
| Agent 37 Cloud C Agent 37 Inc. | auto_sleep checkpoints memory and wakes on any URL request, billing $0.09 a GB-month of disk asleep. The restore takes a few seconds, not 2, and a polling agent's outbound connection does not count as activity, as a launch-thread user asked. source |
| Blaxel Sandboxes C Blaxel, Inc. | Standby keeps memory and processes, resumes in under 25 ms (claimed), bills storage only at $0.20 a GB-month, and a preview request wakes it. Tier 0 and Tier 1 delete sandboxes after 7 or 30 days, so weeks of use need a higher tier. source |
| Celesto (formerly SmolVM) BB Celesto AI, Inc. | Snapshots pause and restore memory, disk and processes and persistent computers keep state, but no idle sleep, wake on request or resume time is documented, and Cloud bills by the running hour. source |
| Cloudflare Sandbox SDK BB Cloudflare, Inc. | Nothing survives a stop unless the app saves a snapshot (filesystem only, public beta, 30 days) and passes it to the next start; processes restart. Issue 102 confirmed nothing is preserved on idle, and a user found no idle auto-destroy. source |
| Daytona BB Daytona Platforms Inc. | Only Linux VM and Windows classes pause with memory and resume; the default container class stops and archives, losing processes. Stopped and paused sandboxes free CPU and memory and bill disk after 5 GiB. source |
| E2B BB FoundryLabs, Inc. | Pause keeps filesystem and memory indefinitely, paused sandboxes are not billed, and auto-resume wakes on traffic, but on_timeout defaults to kill (issue 2056) and issue 1947 lost environment defaults after an envd upgrade at resume. source |
| Modal Sandboxes B Modal Labs, Inc. | No pause: memory snapshots are experimental, expire after 7 days, need the same instance type and cannot be taken during an exec, while filesystem snapshots keep 30 days without processes. source |
| Morph Cloud D Morph Labs, Inc. | Pause preserves memory and processes and charges storage only, with wake_on_ssh and wake-on-request for HTTP, but it needs the Developer plan at $40 a month, and resume time is not stated. source |
| SPACE C Perplexity AI | Pause keeps memory resident and suspend snapshots memory and root filesystem for a durable wake, on your own hardware. No wake on inbound request was found, and wake time is not stated. source |
| Runloop Devboxes B Runloop AI, Inc. | Suspend keeps disk only, processes must restart, suspended devboxes still pay storage, and wake_on_http resumes a tunnel; the service is winding down. source |
| Sandbox0 C Lougao, LLC | Paused sandboxes pay storage only at $0.02 a GiB-month and auto_resume wakes them, but the default pause keeps files and starts fresh processes, memory pause is experimental, and wake-up takes time and needs capacity. source |
| Sprites B Fly.io, Inc. | Disk persists, compute stops on pause, cold storage is $0.000027 a GB-hour, and a request wakes it in 100 to 500 ms warm or 1 to 2 s cold. The warm-to-cold drop is uncontrolled, so processes may not survive an hour. Staff sandboxes run this way. source |
| Together Code Sandbox C Together Computer, Inc. | terminate with memory true saves a snapshot that a new sandbox resumes from with processes intact, but there is no idle sleep or wake on request, and snapshot storage pricing was not found. source |
| Vercel Sandbox BB Vercel Inc. | Persistent sandboxes snapshot the filesystem on stop and any SDK call resumes a new session, with snapshots at $0.08 a GB-month, but memory and processes are not kept and an idle running sandbox bills about $1.05 a day. source |
Doesn't do this job 6
Listings in the category that do something else, with the reason.
| Listing | Why |
|---|---|
| Amazon Bedrock AgentCore Code Interpreter BB Amazon Web Services, Inc. | No pause, resume or snapshot; a session terminates at its timeout and its data is cleaned up. source |
| DeltaBox Lite C DeltaBox (delta-box) | No sleeping machine; sessions are transactions on a local project directory. source |
| Deno Sandbox D Deno Land Inc. | Sandboxes are ephemeral with no pause; a timeout kills the VM. Volumes persist disk in ord only and must be mounted into a new sandbox. source |
| InstaVM D Checksum Labs, Inc. | No pause or sleep; a VM is terminated and recreated from a snapshot, with volumes carrying data. source |
| Microsoft Execution Containers BB Microsoft Corporation | No sleeping machine; each run is a contained process. source |
| Offscreen B Mark Phelps (individual) | The VM clone is deleted after each job; only a per-project cache survives. source |
What stops agents finishing
- timeout kills the sandbox and loses state unless on_timeout is set (E2B)
- container classes stop and archive instead of pausing, so memory is lost (Daytona container class, Vercel snapshots)
- polling agents hold a connection open so the sandbox never sleeps and never stops billing
- persistent VMs or snapshots need a subscription rather than usage billing (Freestyle)
Around it
Before: personal-assistants.new-user, agent-memory.load-context. After: agent-memory.store-context, personal-assistants.reply-in-chat.
Where this comes from
Weight 21 of 100 in Code execution sandboxes: 11 of 53 mentions. Persistence and idle cost are the most repeated themes in the Agent37, Cloudflare and Deno threads, and three GitHub issues (E2B 2056 and 1947, Cloudflare 102) concern what survives a pause.
The same as data: /api/v1/use-cases/code-sandboxes.persistent-agent-machine.json, or the MCP tool get_use_case with id code-sandboxes.persistent-agent-machine. Other jobs in this category: /api/v1/niches?category=code-sandboxes.
For companies
Do agents find, use and choose your tools?
An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.