Temporarily unavailable
Sandbox API is currently paused while we activate an account with our infrastructure provider — API requests will return an error. The section below shows how it will work once enabled; current status is on the /status page.Sandbox API for AI agents
Code Sandboxes for Your AI Agents
Isolated microVMs with a single API call: your agent writes code, the sandbox runs it safely. Start in seconds, pay with crypto — no bank card required.
New users get a starting balance: enough for initial tests without a card
# pip install aineron
from aineron import Sandbox
with Sandbox(template="python") as sbx:
result = sbx.exec(code="print(2 + 2)")
print(result.stdout) # 4Everything You Need for Untrusted Code Execution
VM-Level Isolation
Each sandbox is a separate Firecracker microVM with its own kernel. The agent's code won't reach your data or neighbors.
Start in Seconds
Warm pool of virtual machines: p95 start time under 5 seconds. State is preserved between code runs in one session.
Files, Exec, Logs
Full set: file read/write, command and code execution (Python, Node, bash), background processes, log streaming.
Public URL
Run a web server inside the sandbox — get an HTTPS address accessible from a browser. Demos and previews without deployment.
Pay with crypto
Top up with USDT or TON via Crypto Pay — no bank account, card, or subscription required. Works the same way across the whole platform.
One Key for Everything
The same API key works with chats (GPT-5, Claude), image generation, and sandboxes. The agent is entirely on one platform.
Transparent Pricing, Per Minute
| Size | Resources | Price |
|---|---|---|
| small | 1 vCPU / 1 GiB RAM | 50 credits/min |
| standard | 2 vCPU / 2 GiB RAM | 100 credits/min |
Unused minutes are refunded upon stopping. Launch error — full refund.
Questions and Answers
How is this different from running code in Docker on your own?
Docker containers share the kernel with the host — escaping the container through a kernel vulnerability is possible. Our sandboxes are full-fledged microVMs with a separate kernel, like AWS Lambda. Plus, you don't need to maintain and secure your own servers.
How much does it cost?
50 credits/min for small (1 vCPU / 1 GiB) and 100 credits/min for standard (2 vCPU / 2 GiB). Rounded up to the nearest minute. TTL cost is reserved at creation, unused minutes are refunded upon stopping. Launch error — full refund.
What is this used for?
AI agents that write and execute code (code interpreter); student solution verification in EdTech; safe execution of user scripts in SaaS; quick demos of web applications with a public URL.
Is there an SDK?
Yes, Python SDK: pip install aineron. For other languages — standard REST API with Bearer authorization, curl examples in the documentation.
What are the limitations?
Up to 3 concurrent Sandboxes and 240 minutes per day per account, TTL up to 60 minutes, command output up to 256 KB. Network access from inside is allowlist-based (pypi, npm, etc.); need your own domain — contact support.