FOR
Research agents Coding agents Ops agents SDR agents Marketing agents Trading agents ETL agents Code-review agents Triage agents Support agents

Sandbox for
background agents.

A sandbox for every background agent you ship. Spawn one with a curl, your code runs.

Same sandbox. Persistent or ephemeral.

Spawn an agent Read the docs →

Have questions? Book a call →

Deploy in two steps
# spawn a sandbox
$ curl -X POST https://api.orbcloud.dev/v1/computers \
  -H "Authorization: Bearer $ORB_KEY" \
  -d '{"name":"my-agent","runtime_mb":1024}'
01 HOW IT WORKS

Three primitives.
Spawn. Run. Persist & resume.

Ephemeral when you want it gone. Stateful when you want it to persist. Both, in the same API.

01

Spawn.

One POST. You get a sandbox with persistent state, full Linux, full root, no 15-minute timeouts. Deploy your agent on it.

ORBone curl · sandbox in seconds
OthersSDK install · function wrapper · cold-start
02

Run.

Run for hours, days, weeks. Memory, files, env, LLM context, all survive restarts, redeploys, and crashes. State is the default.

ORBstate survives restarts and crashes
Others15-min timeouts · stateless functions
03

Persist & resume.

Idle sandboxes snapshot to NVMe. Wake on the next event in ~0.6s, same memory, same files, same context. Or DELETE when done.

ORB~0.6s wake · state intact · unlimited session
Othersno sleep · 24h session caps · cold-start tax

Same primitives, one sandbox or many. Read the docs →

02 WHAT YOU GET

Every sandbox
comes with this.

No tiers, no add-ons, no upsells. Same spec on every Orb sandbox.

SPECIFICATION
my-agent
running
  1. 01
    Operating system
    Ubuntu 24.04 LTS · kernel 6.8 · full root
  2. 02
    Storage
    NVMe-backed · persists across sleep, restart, and redeploy
  3. 03
    Network
    Isolated per sandbox · own veth · full outbound · subdomain on {id}.orbcloud.dev
  4. 04
    LLM proxy
    Built-in on 127.0.0.1:8080 · forwards to your provider · buffers across sleep/wake
  5. 05
    Pre-installed
    python 3.12 · node 22 · bun 1.2 · git · curl · jq · tmux · vim · gcc
  6. 06
    Anything else
    Full apt · install whatever your agent needs — Go, Rust, Postgres, Chrome, your stack
04 PRICING

Plans for any scale.
Hobby to enterprise.

Same primitives, same proxy, same metal. The plan just sets the ceiling.

Hobby
$0
For solo devs experimenting.
1 sandbox 100 GB-hours / mo 5 GB disk community support
Sign in with GitHub
Startup
$29/mo
For production background agents.
5 sandboxes 1,000 GB-hours / mo 50 GB disk email support
Start free trial
Scale
$199/mo
For serious workloads.
50 sandboxes 10,000 GB-hours / mo 500 GB disk priority support
Talk to us
Enterprise
Custom
Dedicated capacity, SLA, private cloud.
Unlimited sandboxes dedicated outbound IP custom regions dedicated support
Talk to us
All plans include disk and bandwidth. Overage at $0.005 / GB-hour.
See the rates →
05 USE CASES

Built for long-running,
stateful workloads.

Live case studies first. Generic patterns below.

LIVE

Clawsweeper

Swarm of Codex shards reviewing a repo on a cron. Each replica sleeps between runs, paid only when reviewing.

 LIVE

Code reviewer

Always-on PR reviewer. Sleeps on LLM waits, resumes mid-diff when the response lands.

 LIVE

Antibiotic scientist

Overnight literature sweep, protein-structure runs. Computers alive for days.

 LIVE

Exoplanet hunter

Cycles NASA TESS target stars, scoring candidate signals between sleeps. Computers alive for weeks.
also common

Coding agents

Claude Code, Codex, Aider. Give each their own computer. Deploy hundreds in parallel.

Orchestrators

CrewAI, LangChain, Composio pipelines. Each gets its own isolated computer.

Research agents

Overnight crawling, summarization, analysis. Computers run 24/7 for pennies.

Browser agents

Playwright, browser-use at scale. Dedicated IP per computer.

Sales & outreach

Multi-day campaigns. Agents that personalize, track, and follow up over days.

Support agents

Multi-turn customer support with persistent memory. Resume the same thread weeks later, context intact.

06 QUESTIONS

Common questions.

What is a background agent?+
An AI agent that runs independently of a user session, code reviewers, research agents, SDR campaigns, monitoring, pipelines. Unlike foreground chat, background agents need a place to live across hours, days, or weeks.
How does Orb compare to E2B, Daytona, Modal, or Blaxel?+
Orb is built specifically for background agents, long-lived, persistent, idle-aware. E2B and Daytona target short-lived AI code execution sandboxes. Modal is serverless functions with cold starts. Blaxel keeps sandboxes on always-warm standby. Orb is the only one where idle sandboxes truly sleep, snapshot to NVMe, and wake in ~0.6s with full state.
Can I run Claude Code, Aider, Cline, or Codex in the background?+
Yes. Templates exist for the Claude Code SDK and OpenClaw. For anything else: any agent that runs on Linux runs on Orb. Full Ubuntu, full root, install whatever your agent needs.
What happens to my agent when it's idle?+
It's snapshotted to NVMe. State (memory, files, LLM context) is preserved. Wakes on the next event in ~0.6s.
Can I run an agent for weeks without losing state?+
Yes. State survives sleeps, restarts, and redeploys. Crashes resume from the last snapshot. Verified across thousands of production operations, zero losses.
What about networking, IPs, outbound traffic?+
Every sandbox has its own isolated network, own veth, nftables-routed. Outbound HTTP/HTTPS to anywhere. Dedicated outbound IP available on Enterprise or by request.
What if I want to run a one-shot task and tear it down?+
DELETE /v1/computers/$ID when done. Fire-and-forget works fine. You only pay for what ran.
Where do my sandboxes actually run?+
Bare-metal hardware in Hetzner (EU). No hyperscaler markup, no surprise egress fees. Custom regions available on Enterprise.

Ship agents to production
this week.

Solo or small team? Sign in with GitHub and start. Bigger team, custom needs, or want to talk first? Drop your email below.

Read the docs