Give your agent a real machine — isolated, persistent,
and production-ready. Full Linux sandboxes controlled
through one SDK, with tools that extend what it can do.
Every sandbox is a full Linux environment — persistent or short-lived. Deploy your agent to live inside one, or let your agent spin them up as disposable tools.
Deploy your agent into a persistent sandbox. Full Linux, root access, real networking. It runs as if it owns the machine — because it does. Custom domains, SSL, and monitoring built in.
Spin up sandboxes on demand — your agent executes code, reads and writes files, runs tests. When it's done, the sandbox disappears. Controlled through SDK or MCP.
Same infrastructure. Same SDK. You choose what the sandbox does.
Every primitive your agent needs — grouped by what it enables, not what it is.
Complete networking stack with exposed ports and firewall rules. Agents communicate, serve, and connect.
Every sandbox is isolated. Process, network, and filesystem separation per agent and tenant.
WebSocket connections stream output in real-time. Watch agents execute live.
Full root SSH into running sandboxes. Debug, inspect, or take manual control.
Agents navigate sites, fill forms, extract data, and capture screenshots with headless Chrome.
Agents search the web and extract content — pages, PDFs, structured data.
Agents install packages, run services, and manage processes like a real system. Python, Node.js, Bash, Go — any runtime, full stdin/stdout streaming.
A full Linux filesystem with persistent volumes. Agents read, write, and organize files across sessions — nothing is lost between runs.
One call turns any sandbox into a live production URL. Custom domains, auto SSL, and edge routing — your agent ships what it builds.
Semantic search. Batch file patching. Environment snapshots. Operations that would take days to build with shell scripts— exposed as MCP resources and tool definitions your agent can call instantly.
Execute any command — shell, Python, Node, Go. Stream stdout in real-time.
result = await sandbox.exec("python train.py")Code it or click it. Every feature is available as MCP resources, tool definitions, or SDK calls for your agent— and through a fully managed dashboard when you need manual control. Same power, your choice.
No SSH, no bash scripts. Semantic search, diff patching, snapshot rollback — capabilities you'd spend days building with shell commands. Here, they're one tool call away.
result = await sandbox.exec("python train.py")await sandbox.files.patch([{ path: "/src/api.ts", changes: [...] }])const files = await sandbox.files.search("auth middleware")const snap = await sandbox.snapshot("before-deploy")await sandbox.publish({ port: 3000, domain: "app.oblien.app" })const matches = await sandbox.grep({ pattern: "async.*fetch", semantic: true })Real infrastructure for real use cases. Each agent gets a full machine to work with.
Run any agent framework on isolated sandboxes
Run any open-source agent like OpenClaw, AutoGPT, or your own — each gets a secure, isolated cloud sandbox with full power. Persistent memory, unrestricted networking, and real compute. Your agent runs 24/7 without touching your local machine.
Full SDKs for Python, JavaScript, and Go — or deploy agents straight from the dashboard. Your call.
from oblien import Sandbox
# Create a persistent sandbox for your agent
sandbox = Sandbox.create(persistent=True, memory="4GB")
# Your agent runs code, browses, stores files
result = sandbox.exec("python agent.py --task research")
sandbox.fs.write("/data/report.md", result.output)
# Deploy to a live URL when ready
url = sandbox.publish(port=3000, domain="my-agent.oblien.app")
# Or just SSH in to inspect
sandbox.ssh() # → root@sandbox-3d1b:~$Deploy any open-source agent in seconds — no code required.
Security is hard. Isolation is hard. Multi-tenant billing is hard. Per-user tracking is hard. Everything's built in — so you ship agents, not infrastructure.
.create().config.namespace.credits.tools.*.publish()Start free, scale as you grow. No hidden fees, no surprise bills. Every plan includes full access to all APIs and tools.
For experimenting and building prototypes
For shipping production agents
For teams running agents at scale
For organizations with custom needs
Namespaces, credit metering, quota enforcement, and billing primitives — all built in. Ship a multi-tenant product without building the infrastructure layer yourself.
Every tenant gets a namespace. Sandboxes, usage, and billing are scoped and isolated automatically — no cross-contamination.
Compute, storage, tools — everything maps to credits. Accurate down to the second. Set prices, margins, and thresholds your way.
Sandbox hours, concurrent instances, API rate limits — all enforced at the platform level. Hit the limit, paused instantly.
Bring your own auth. Pass any identifier — our namespace layer handles isolation, metering, and billing from there.
const ns = await oblien.namespace.create({
tenantId: "org_acme",
plan: "pro",
limits: {
sandboxHours: 100,
concurrency: 5,
credits: 5000
}
});
// All usage auto-metered
const sb = await ns.sandbox.create();
// → credits deducted per second
// → quotas enforced at the edgeEvery account starts with free credits — enough to prototype, test, and ship your first integration. No credit card required. When you scale, credits refill automatically based on your plan.
Machines, tools, networking, and deployment. One platform. Start free, scale to production.
Get updates on new primitives, sandbox features, and best practices for building production-grade AI agents.
No spam. Unsubscribe at any time.