./ai-sandbox.sh

Isolated GPU sandboxes for AI teams. Built in LATAM, for LATAM.

New

Spin up secure, isolated GPU and CPU sandboxes in seconds — for reinforcement learning, agent tool use, model evaluation, and inference. Your data never leaves the region. Your users see sub-30 ms latency from São Paulo, Bogotá, Santiago, and CDMX.

AI Sandbox is the execution layer your agentic and RL workloads need — isolated, regional, sovereign. Built on EdgeUno bare metal in LATAM data centers, exposed through a Python SDK that takes four lines to get from pip install to a running GPU job.

The same pattern your team already knows from US-hosted sandboxes — except the GPU is in São Paulo, the data never crosses a border, and the support team speaks your language.

./why-ai-sandbox.sh

What you get.

  • /shield-lock

    Hard isolation by default

    Every sandbox runs in its own hardware-virtualized environment. A runaway agent, a memory spike, or a malicious tool call in one sandbox can never touch another.

  • /flag

    LATAM-resident, LGPD-aligned

    Compute and storage stay in-region. Brazilian data stays in Brazil. Mexican data stays in Mexico. Sovereignty controls built in, not bolted on.

  • /zap

    Cold start in seconds

    Pre-warmed pools across our LATAM PoPs. pip install, four lines of Python, your sandbox is running.

  • /layers

    Massively parallel

    Spin up hundreds of CPU sandboxes per training step for RL rollouts. Run thousands of evaluation jobs in parallel without provisioning a single node.

  • /globe

    Low latency from LATAM users

    24 PoPs across Brazil, Mexico, Colombia, Chile, Argentina, Peru. Build agentic apps your São Paulo and CDMX users can actually feel.

  • /terminal

    Python SDK + REST + CLI

    Drop-in Python client, REST API, and CLI. Compatible patterns with the wider ecosystem — minimal lift to migrate workloads from US-hosted sandboxes.

./differentials.sh

Why AI Sandbox on EdgeUno.

Every product shares the same LATAM-resident network. Here's what that means for you specifically.

./how-it-works.sh

How it works.

  1. 01

    Request access

    30-second form. Our team confirms inventory in your target region within one business day.

  2. 02

    Get credentials

    API key, region selection (BR / MX / CO / CL / AR / PE), and SDK install instructions.

  3. 03

    Spin up your first sandbox

    Four lines of Python. Your first GPU sandbox is running in your chosen LATAM region.

  4. 04

    Scale

    Move from one sandbox to thousands. Talk to us about reserved capacity and committed-use discounts.

./use-cases.sh

Built for.

  • LATAM startups training on LATAM data

    Fine-tune Portuguese and Spanish-language models on customer data that legally cannot leave the country.

  • Agentic AI for regional apps

    Run code-execution agents for fintech, e-commerce, and logistics apps with sub-30 ms latency from your end users.

  • Reinforcement learning rollouts

    Spawn thousands of isolated environments per training step. Pay per sandbox-second, not per reserved node.

  • Model evaluation at scale

    Run continuous eval suites in parallel sandboxes. Capture every trace, every artifact, every run.

  • Research institutions and universities

    Sovereign GPU access for LATAM research groups without exporting datasets to US-hosted clouds.

./quickstart.sh

Four lines of python.

from edgeuno_sandbox import Sandbox

with Sandbox.run(gpu="a10", region="br-sao") as sandbox:
    result = sandbox.exec(["python", "train.py"]).result()
    print(result.stdout)

          

./gpu-options.sh

GPU inventory.

Inventory subject to confirmation per region — see M11 inventory pass before publishing rates.

Available now

  • NVIDIA T4

    Inference, lightweight fine-tuning

  • NVIDIA A10 / L4

    Inference, mid-size fine-tuning, agent workloads

  • NVIDIA A100 40/80GB

    Training, large-model inference, RL rollouts

On roadmap

  • NVIDIA H100

    On roadmap

./pricing.sh

Per-second billing on CPU and GPU

Preview pricing. Committed-use discounts on request.