Modeva Documentation & User Guide
Modeva bridges RunPod, Vast.ai, and Novita AI into one seamless, unified control plane. Compare live spot rates across clouds, allocate persistent physical storage volumes, auto-snipe restocked GPUs, or chat with our autonomous AI agent to deploy compute in seconds.
Live Price Comparison
Real-time spot and on-demand pricing across RunPod, Vast.ai, and Novita. Save up to 70% on compute by picking the best market rate.
GPU Sniper Bot
Can't find an RTX 4090 or H100 in stock? Arm a sniper. It scans inventory every 60s and automatically claims the instance the instant it restocks.
Your Own Keys
You pay providers directly with zero middlemen fee. Your API keys are encrypted client-side and saved with AES-256-GCM.
# 1. What is Modeva? (The Simple Explanation)
Think of Modeva like Google Flights or Skyscanner, but for AI GPUs and cloud compute.
Normally, if you want to rent an RTX 4090, A100, or H100, you would have to open three different tabs, log into RunPod, Vast.ai, and Novita AI separately, check whose datacenter has free capacity, and manually compare prices.
Modeva unifies all of them in one place:
- One single catalog of live available GPUs from all 3 providers.
- Live spot rates updated every few minutes so you never overpay.
- 1-Click deployment directly into your own cloud accounts.
- Zero extra fees or markups — you pay the cloud providers directly.
# 2. Step-by-Step Guide: Launching Your First GPU
Getting your first containerized GPU pod up and running takes less than 2 minutes:
Head to the API Keys page. Paste your personal API key from RunPod, Vast.ai, or Novita AI. Your keys are immediately encrypted using AES-256-GCM before being stored.
Go to the GPU Catalog. Filter by GPU architecture (e.g. RTX 4090, A100 SXM, H100), VRAM (24GB, 48GB, 80GB), or max price. You can toggle "Deployable Only" to see machines ready for immediate launch.
Click DEPLOY → on any available card. Choose your Docker container image (default PyTorch 2.4 + CUDA 12.4), specify container disk size, select an optional persistent network volume, and hit Deploy Pod Instance.
Visit the Pods tab to see your running instances. Click COPY SSH to get your direct command, e.g. ssh root@ip -p port, or connect directly to exposed ports (Jupyter, ComfyUI, Ollama).
# 3. API Keys & Security Vault
Modeva is built from the ground up as a zero-trust, non-custodial platform. We never ask for credit cards or store your cloud passwords.
🔐 AES-256-GCM Encryption
When you save a key in the API Keys tab, it is encrypted on the server with AES-256-GCM using an isolated user key and unique cryptographic nonce. Plain-text keys are never stored in databases.
💳 Direct Billing
Every container spun up through Modeva is allocated under your personal RunPod, Vast.ai, or Novita account. You pay your exact provider invoice directly, receiving all native provider volume credits and discounts.
- RunPod: Account Settings → API Keys → Create New Key
- Vast.ai: Account → API Key → Copy
- Novita AI: User Settings → API Keys → Add Key
# 4. GPU Catalog & Live Arbitrage
Different cloud providers price the exact same GPU models differently based on real-time datacenter demand. Modeva tracks these price discrepancies so you can arbitrage the market:
| GPU Architecture | VRAM | Best Spot Rate | Best On-Demand | Recommended For |
|---|---|---|---|---|
| NVIDIA RTX 4090 | 24 GB GDDR6X | $0.19 - $0.34 /hr | $0.44 - $0.59 /hr | LoRA training, ComfyUI, LLM Inference (8B-14B) |
| NVIDIA L40S | 48 GB GDDR6 | $0.55 - $0.79 /hr | $0.85 - $1.05 /hr | High VRAM fine-tuning (32B models), vLLM Serving |
| NVIDIA A100 SXM4 | 80 GB HBM2e | $1.10 - $1.89 /hr | $1.89 - $2.20 /hr | Deep learning training, 70B Quantized LLMs |
| NVIDIA H100 SXM5 | 80 GB HBM3 | $1.99 - $2.49 /hr | $2.99 - $3.49 /hr | FP8 Ultra-fast training, Enterprise Multi-node |
# 5. Spot vs On-Demand (Postpaid) Explained Simply
When launching a pod, you will often see options for Spot (Community) vs On-Demand (Secure/Postpaid). Here is how to choose:
- • Cloud providers sell their idle/spare capacity at a heavy discount.
- • Trade-off: If another customer is willing to pay full price, your pod could be paused or preempted.
- • Best for: Checkpointed model training, batch rendering, experimental scripts, and non-critical workloads.
- • Dedicated compute reserved exclusively for you.
- • Guaranteed: Will never be interrupted or shut down until you stop or delete it.
- • Best for: Live inference APIs, production web applications, long uninterrupted training runs.
6. GPU Sniper: Automatic Restock Hunter
NEW FEATUREHigh-demand GPUs (like NVIDIA RTX 4090s or H100s) frequently sell out on RunPod, Vast, and Novita within seconds. Instead of sitting and refreshing the webpage every 5 minutes, arm a GPU Sniper.
🎯 How the Sniper Works
Configure Target
Choose GPU model (e.g. RTX 4090), max hourly price (e.g. $0.35/hr), and preferred provider.
60s Background Polling
Modeva's background daemon monitors live spot feeds from RunPod, Vast.ai, and Novita every minute.
Instant Auto-Launch
The exact millisecond an instance frees up within your price ceiling, Modeva triggers create_pod.
Instant Alert
Receive an instant email or webhook ping containing your pod IP and SSH command.
⚠️ Safety First: Dead Man's Switch
What if you set a sniper for 3 AM, fall asleep, and forget about it? You wouldn't want an instance running and billing you for days unmonitored.
Every sniper comes with a configurable Dead Man's Switch (Max Window): 2 hours, 6 hours, 12 hours, or 24 hours. If an instance isn't found within your chosen window, the sniper automatically disarms itself. Zero surprise bills.
# 7. Persistent Storage & Physical Volumes Engine
Containers are ephemeral by default — if you delete a pod, any downloaded weights (like a 40GB LLM or checkpoint files) are lost. Persistent volumes keep your data intact permanently across pod restarts.
RunPod Datacenter Volumes
RunPod network volumes live at the datacenter cluster level (e.g. EU-RO-1, US-CA-1). Any compatible GPU in that datacenter can mount your storage seamlessly at /workspace.
Vast.ai Host Machine Matching
Vast.ai volumes live on specific physical hardware (machine_id). Modeva filters 5,000 live offers to show you only GPUs on that exact physical machine, ensuring volume attachments never fail.
# 8. Managing Pods & SSH Access
Once your instance is deployed, manage its lifecycle from the Pods dashboard:
Direct encrypted shell terminal into your pod container.
ssh root@<ip> -p <port>Click directly on exposed port links (8888 for Jupyter, 7860 for Gradio/ComfyUI).
Pause compute to stop GPU billing while keeping your volume storage intact.
9. Modeva Autonomous Cloud AI Agent
GROQ LLMNeed compute fast? You can talk to Modeva AI directly in plain English. Just click the ⚡ badge in the bottom right corner of any page.
The agent uses recursive planning: it checks live catalog stock, matches datacenter regions, resolves physical volumes, and generates a structured confirmation plan before executing any cloud deployment.
# 10. REST API & cURL Reference
Build your own automation, CLI tools, or CI/CD pipelines using Modeva's REST API:
curl -X GET "https://api.modeva.ai/gpu_catalog?available_only=true" \
-H "Accept: application/json"curl -X POST "https://api.modeva.ai/api/v1/snipes" \
-H "Authorization: Bearer <CLERK_TOKEN>" \
-H "Content-Type: application/json" \
-d '{
"target_gpu": "RTX 4090",
"target_provider": "runpod",
"max_price": 0.35,
"auto_deploy": true,
"max_window_hours": 6
}'curl -X POST "https://api.modeva.ai/runpod/create_pod?name=my-pod&gpu_id=NVIDIA%20GeForce%20RTX%204090&image_name=runpod/pytorch:2.4.0-py3.11-cuda12.4.1-devel-ubuntu22.04" \
-H "Authorization: Bearer <CLERK_TOKEN>"