Now in public beta

GPU compute for
AI agents.

Your AI coding agent submits Python. We execute on RTX 4090 GPUs and return artifacts via MCP. Per-second billing, reproducible to the seed, built for research-grade workflows.

Start Free with $5 CreditTry the Playground
17s LoRA fine-tune·$0.024 per image·164 TFLOPS FP16·MCP-native
SCROLL↓

Fine-tune a LoRA in 17 seconds. Generate 4K images for under a penny. Embed an entire arXiv corpus for $0.31. Run any Python on any GPU through one MCP call.

Reproducible to the random seed. Every run has a permanent URL. Built for AI engineers who care about rigor.

Your code, delivered.

Submit any Python script via MCP tool call or REST API. We handle provisioning, execution, and artifact return. No containers to build, no instances to manage.

Try it

Per-second billing

Pay only for the seconds your code runs. No idle charges, no minimum commitments. Your vault balance decrements in real-time.

Learn more

RTX 4090 GPUs

Every run executes on NVIDIA RTX 4090 hardware. 16,384 CUDA cores, 24GB VRAM, PCIe 4.0. Ideal for inference, fine-tuning, and batch processing.

Learn more

MCP native

First-class Model Context Protocol support. Your AI agent calls run_on_gpu as a tool, receives structured results with stdout, stderr, and artifact URLs.

Try it

Code never stored

Your source code is never persisted in our database. We store only a SHA-256 hash for audit. Output artifacts are encrypted in R2 with time-limited access.

Learn more

Team workspaces

Organize agents by project, share vault balances, and audit every run across your team. Role-based access with admin, member, and viewer tiers.

Learn more
main.pyGenerate a 1024x1024 image with Stable Diffusion XL
Loading editor...
output

Click "Run on GPU" to execute

How it works

Three steps. Zero infrastructure.

01

Submit code

Your AI agent calls the run_on_gpu MCP tool with a Python script and GPU requirements.

{
  "tool": "run_on_gpu",
  "arguments": {
    "code": "import torch\n...",
    "gpu_type": "RTX_4090",
    "timeout": 300,
    "packages": ["torch", "diffusers"]
  }
}
02

We execute

Code runs on a warm GPU worker. Dependencies install from cache. Stdout streams in real-time.

Agent: Run this SDXL generation script.

Volspan: Provisioning RTX 4090...
Volspan: Installing packages (cached)
Volspan: Executing main.py
Volspan: ████████████ 100%
Volspan: Run complete. 4.7s, $0.024
03

Get results

Structured response with stdout, stderr, artifacts (R2 signed URLs), cost, and duration.

{
  "status": "completed",
  "duration_ms": 4700,
  "cost_cents": 2,
  "stdout": "Image saved: 1024x1024 PNG...",
  "stderr": "",
  "artifacts": [
    {
      "path": "/output/generated.png",
      "url": "https://r2.volspan.dev/...",
      "size_bytes": 2048576
    }
  ]
}

Simple, transparent pricing

Pay for what you use. All plans include per-second GPU billing.

Hobby
$0
  • ✓$5 free credits/mo
  • ✓1 agent
  • ✓RTX 4090
  • ✓5 min timeout
  • ✓Community support
Get Started
Pro
$29/mo
  • ✓$50 credits included
  • ✓5 agents
  • ✓RTX 4090
  • ✓15 min timeout
  • ✓Priority queue
  • ✓Email support
Start Free Trial
Team
$99/mo
  • ✓$200 credits included
  • ✓25 agents
  • ✓RTX 4090 + A100
  • ✓30 min timeout
  • ✓Workspaces
  • ✓Priority support
  • ✓Audit logs
Start Free Trial
Business
$299/mo
  • ✓$600 credits included
  • ✓100 agents
  • ✓All GPU types
  • ✓60 min timeout
  • ✓SSO/SAML
  • ✓Dedicated support
  • ✓SLA 99.9%
Contact Sales
Scale
$799/mo
  • ✓$2000 credits included
  • ✓Unlimited agents
  • ✓All GPU types
  • ✓No timeout limit
  • ✓Custom VPC
  • ✓Dedicated account mgr
  • ✓SLA 99.95%
Contact Sales
Enterprise
Custom
  • ✓Volume discounts
  • ✓Unlimited everything
  • ✓Private clusters
  • ✓On-prem option
  • ✓Custom contracts
  • ✓24/7 phone support
  • ✓SLA 99.99%
Talk to Us

Academic pricing available: 50% off any tier with .edu email. Apply →

Ship AI features faster.
Start in seconds.

Get Started Free

No credit card required · Free tier never expires