Compute · Nirvana Agent Sandboxes

Fast, Isolated
Sandboxes Built for
Agent Economics.

Fast, isolated, persistent compute for running autonomous agents at scale. Create, run, pause, resume, delete. On Nirvana's own cloud.

Read the DocsTalk to Us
ephemeral~2.5s boot
Pure compute, no disk
persistent6-7s first boot
/workspace survives
Agent Sandboxes illustration
Nirvana Agent Sandboxes

A Sealed Place to Run Agent Code.
Fast, Isolated, Persistent.

01
Speed

Live in Seconds

Ephemeral sandboxes boot in ~2.5 seconds, persistent in 6-7. Pause, resume, and delete are instant, so agents restart mid-task with zero wait.

02
Isolation

Two Layers of Isolation

Container walls around each sandbox, VM walls around each node, and a dedicated NKS cluster per customer. Your sandboxes never share a kernel, or anything else, with another company.

03
Persistence

Persistent Storage

Everything the agent writes to /workspace is backed by our own Accelerated Block Storage (ABS) and survives pause and resume. Pause instantly to stop paying for compute; resume instantly with every file exactly where it was.

04
Compatibility

Six SDKs, One API

Create, run, pause, resume, delete. One endpoint manages the loop, one gateway gets you inside. Available in Python, TypeScript, Go, Java, Kotlin, C#.

05
Infrastructure

Purpose-Built Infrastructure

No hyperscalers, no third-party cloud. AMD EPYC 9354P (32C/64T, 3.25-3.8 GHz), DDR5, hard anti-affinity: a host can fail and the cluster keeps operating with zero downtime.

06
Pricing

Flat Pricing to Run Agents at Scale

Built from Nirvana Cloud's flat rates, 20% cheaper than leading sandbox providers. ABS includes 20,000 sustained IOPS with no surcharge and no overprovisioning.

With a Disk, or Without //

Sandbox Options

Both run on the same isolated substrate. The only question is whether the agent's work needs to outlive the sandbox.

Option 01 //
ephemeral
Pure Compute, No Disk.

Boots, runs its task, and leaves nothing behind. One clean environment per job, gone the moment it's done.

boot ~2.5s · no disk · gone on delete
Option 02 //
persistent
Checkpoint and Continue.

A dedicated volume backed by our own ABS, mounted at /workspace. Pause instantly to stop paying for compute; resume instantly with every file exactly where it was.

first boot 6-7s · /workspace on ABS · pause/resume instant
Operations //

Every operation in the API.

create (ephemeral)~2.5s
Endpoint: APIClean environment
create (persistent)6-7s
Endpoint: API/workspace mounted from ABS
pauseInstant
Endpoint: APICompute released, files held
resumeInstant
Endpoint: APIFiles intact in /workspace
deleteInstant
Endpoint: APINo residue
exec / filesLive
Endpoint: GatewayRun commands, read/write files
~2.5s
Ephemeral Boot
instant
Pause and Resume
20,000
Sustained IOPS Included
6
Native SDKs
1
Dedicated Cluster per Customer

Powering AI, blockchain, and
databases

Talk to Sales