# ParalonCloud

> ParalonCloud is a decentralized AI compute network (DePIN). Rent on-demand NVIDIA GPUs by the minute, run LLM inference through an OpenAI-compatible API, or monetize an idle GPU by sharing it and earning real USDC. Instead of one company owning the hardware, independent providers worldwide share their GPUs — and that pooled compute powers everything from a one-line model API to full GPU rentals.

## Summary

ParalonCloud has three sides that run on one network:

1. Rent GPUs — spin up an NVIDIA GPU in your browser in under a minute, billed per minute.
2. AI Model API — call open models for text, vision, image generation and text embeddings through an OpenAI-compatible API; a free key carries a 250,000-token trial, after that it is priced per token from prepaid credits.
3. Provide compute — turn an idle GPU into income, earning real USDC every minute it's rented, plus points in an incentivized testnet.

## Platform URL

https://paraloncloud.com

## What is a decentralized GPU network (DePIN)?

DePIN (Decentralized Physical Infrastructure Network) is a model where individuals contribute physical hardware to a shared network. On ParalonCloud, GPU owners around the world plug in their machines and earn when their compute is used, while developers rent that compute on demand — no centralized cloud, no long-term contracts, pay only for what you use.

## Monetize Your GPU — Make Money With an Idle GPU

Have an idle or underused GPU? Put it to work on ParalonCloud and earn. This is one of the simplest ways to monetize a gaming or workstation GPU, or a fleet of data-center cards.

How you earn:

- Real USDC from rentals. When someone rents your node, you earn real USDC — credited to your balance every minute the node is rented (per-minute billing), so you start earning from the first minute. Providers keep 80% of the rental (80/20 split: 80% provider, 20% platform fee).
- Instant, on-platform earnings. Earnings accrue in real time as your node is rented — you watch the balance grow live in your dashboard, no invoicing or waiting to see what you made.
- Withdraw to your wallet. Cash out in USDC on Solana to a Phantom or Solflare wallet. No withdrawal fees. Or convert earnings into platform credits at 1 USDC = 1 credit to spend on your own rentals or inference.
- Incentivized testnet (Season 0). Separately, you earn points just for keeping your node online and verified — no rental required. Points = VRAM × time (1 GB of VRAM = 1 point per minute; multiple GPUs stack). Points rank you on the public leaderboard during Season 0, the program that rewards early providers.

Getting started as a provider:

- Requirements: an NVIDIA GPU (RTX 3090/4090/5090 through A100/H100/H200), up-to-date NVIDIA drivers, Docker, and the NVIDIA Container Toolkit. Linux or Windows (WSL2).
- Connect with one command: run a single Docker command on your machine. The agent connects out through a secure tunnel — no port forwarding and no public IP required.
- You stay in control: accept rentals when you want, stop hosting anytime.

Keywords: monetize GPU, make money with GPU, earn USDC with GPU, idle GPU income, rent out your GPU, GPU passive income, sell GPU compute, DePIN rewards, GPU mining alternative.

## Rent GPUs — On-Demand NVIDIA Compute for AI & Developers

Need a GPU now? Rent on-demand NVIDIA GPUs on ParalonCloud and get a full environment in your browser in under a minute. Built for AI engineers, ML researchers, and developers who need GPU power without buying hardware or signing a cloud contract.

- Billed per minute. Pay as you go, billed per minute (rates shown per hour). A short 1-hour balance is required up front to open a session; you're only charged for the minutes you actually use. Stop anytime.
- Full Jupyter Lab in the browser. Every rental gives you a Jupyter Lab environment — Python, a terminal, a file browser, and `nvidia-smi`. `pip install`, run training, fine-tune, serve a model.
- Real NVIDIA hardware. From consumer RTX 4090s to data-center A100/H100/H200.
- Isolated and secure. Each rental runs in its own container, reachable through a secure tunnel.

Great for: AI model training, LLM fine-tuning, machine-learning inference, image/video generation (Stable Diffusion, etc.), running notebooks, deep-learning experiments, and any general GPU workload — without owning the hardware.

Keywords: rent GPU, rent a GPU online, on-demand GPU, cloud GPU, GPU for AI, GPU for machine learning, rent NVIDIA GPU, RTX 4090 rental, A100 rental, H100 rental, GPU cloud for developers, cheap GPU rental, Jupyter GPU.

## Rental API — Rent GPUs Programmatically

Rent GPU servers over HTTP, from code — the programmatic version of browser GPU rental. Browse available GPUs with live prices, start a rental, poll for its connection URL, and stop it, all with a simple REST API. It uses the same `prlc_` key as the Inference API, gated by a per-key `rental` scope (opt-in, because rentals spend credits — a leaked key can't drain your balance unless you grant it).

Base URL: https://paraloncloud.com/api/v1

Endpoints:
- GET /api/v1/gpus — list rentable GPU nodes with price per hour, GPU model, VRAM, CUDA compute capability, and country. You rent by node_id.
- GET /api/v1/balance — your credit balance
- POST /api/v1/rentals — start a rental (asynchronous; returns a rental_id with status "pending"). Body: node_id, type ("jupyter"), optional hours (auto-stop after N hours), optional name. Send an Idempotency-Key header so a retried call doesn't start a second GPU.
- GET /api/v1/rentals — your active rentals (add ?status=all for the full history)
- GET /api/v1/rentals/{id} — poll until connection_pending is false, then read the connection URL + token
- DELETE /api/v1/rentals/{id} — stop the GPU and stop billing (idempotent)

Renting is asynchronous: create returns immediately with a rental_id, then you poll GET /rentals/{id} until it's "running" to get the connection details. Rentals bill per minute; set hours for a time-limited rental that auto-stops even if your script crashes. Each key has a configurable max_active_rentals limit as a safety net.

cURL example:
```bash
# 1. find a GPU
curl https://paraloncloud.com/api/v1/gpus -H "Authorization: Bearer prlc_your_key_here"

# 2. rent it (async — returns a rental_id; poll GET /rentals/{id} for the URL)
curl -X POST https://paraloncloud.com/api/v1/rentals \
  -H "Authorization: Bearer prlc_your_key_here" \
  -H "Idempotency-Key: my-run-001" \
  -H "Content-Type: application/json" \
  -d '{"node_id": "<from /gpus>", "type": "jupyter", "hours": 2}'
```

Authentication: same `prlc_` key as inference; enable the `rental` scope on the key in the Console. Create and manage keys at https://paraloncloud.com/console.

MCP server: rent GPUs from inside Claude Code, Claude Desktop, or Cursor with the ParalonCloud Rentals MCP server (npm package @paraloncloud/mcp-rentals, source github.com/ParalonCloud/mcp-rentals). It exposes the rental API as MCP tools (list_gpus, get_balance, create_rental, get_rental, list_rentals, destroy_rental). Config: command `npx -y @paraloncloud/mcp-rentals` with env PARALON_API_KEY set to a rental-scoped key. Docs: https://paraloncloud.com/docs/rental-api/mcp.

Great for: automated GPU provisioning, CI/CD GPU jobs, agent-driven compute (an AI agent that rents its own GPU), batch training pipelines, spinning up GPUs on demand from scripts, and orchestration tools that rent and release GPUs programmatically.

Keywords: GPU rental API, rent GPU programmatically, GPU cloud API, on-demand GPU API, provision GPU from code, GPU API for AI agents, automated GPU rental, REST GPU API, rent NVIDIA GPU with API, RunPod API alternative, Vast.ai API alternative.

## AI Model API — OpenAI-Compatible Text, Vision, Image and Embeddings

Run open AI models over HTTP — no GPU to provision, no model to download. The API is OpenAI-compatible, so any tool or SDK that speaks the OpenAI API works by changing two lines: the base URL and the key. Requests are served on distributed GPU nodes across the network.

One `prlc_` key covers every model on the network: text and code, vision (image input), image generation and text embeddings. Live today: Qwen 3.8 27B with vision (qwen3.8-27b), Qwen3 3B (qwen3-3b), Z-Image Turbo (z-image-turbo) for text-to-image, and Multilingual E5 Small (multilingual-e5-small) for embeddings. The same key works for any model added later, with no change on your side. Speech endpoints exist but no speech model is currently serving.

- Priced per token, from prepaid credits. A free key carries a 250,000-token trial; a premium key pays list price: qwen3.8-27b $0.12 per 1M input and $1.70 per 1M output tokens, qwen3-3b $0.05 / $0.20, z-image-turbo $0.003 per image, multilingual-e5-small $0.02 per 1M input tokens. Usage is reported on every response (`prompt_tokens`, `completion_tokens`, `total_tokens`).
- Limits: free keys 20 requests per minute and 2 in flight; premium keys 300 per minute and 16 in flight.
- Drop-in OpenAI SDK replacement. Point the official OpenAI SDK at ParalonCloud.
- Streaming and non-streaming. Set `stream: true` for token-by-token responses.
- Try without code. Use the Playground to chat with models in your browser, or the Console to create keys and track usage.

Base URL: https://paraloncloud.com/v1

Endpoints:
- GET /v1/models — list available models (read this for the live catalog; available models change as provider nodes come online)
- POST /v1/chat/completions — chat and vision completions (OpenAI compatible; pass an image in the message content for vision models)
- POST /v1/responses — the OpenAI Responses API shape (input items, instructions, function tools, text.format JSON schema, streaming events) on the same chat models; no stored conversation state, no built-in tools
- POST /v1/files, POST /v1/batches, GET /v1/batches/{id}, GET /v1/files/{id}/content — the OpenAI Batch API: upload a JSONL file of chat, embeddings or Responses requests, results within 24 hours at 50% of the token price, run on idle capacity without counting against the key's rate limits
- POST /v1/images/generations — text-to-image (OpenAI Images API shape; returns `b64_json`)
- POST /v1/embeddings — text embeddings (OpenAI Embeddings API shape; one string or an array of up to 256; 384-dimension vectors from multilingual-e5-small, served on CPU nodes of the network)
- POST /v1/classify — typed decisions from a classification model: a label and calibrated probabilities per input, no generated text (input: a string or up to 256 strings, or pairs of {premise, hypothesis} for NLI models; used for reranking, grading, guardrails and routing; models listed by GET /v1/models with "type": "classification"; billed per input token)
- POST /v1/audio/transcriptions — speech-to-text
- POST /v1/audio/speech — text-to-speech

Python example:
```python
from openai import OpenAI

client = OpenAI(
    api_key="prlc_your_key_here",
    base_url="https://paraloncloud.com/v1",
)

response = client.chat.completions.create(
    model="qwen3-8b",
    messages=[{"role": "user", "content": "Hello!"}],
)

print(response.choices[0].message.content)
```

cURL example:
```bash
curl https://paraloncloud.com/v1/chat/completions \
  -H "Authorization: Bearer prlc_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3-8b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'
```

Authentication: API keys start with the `prlc_` prefix; create them in the Console and send them as a Bearer token in the Authorization header.

Models: ParalonCloud serves open-weight models. Live today: **Qwen 3.8 27B with vision** (qwen3.8-27b, 32k context, accepts image input), **Qwen3 3B** (qwen3-3b, 16k context, served on Apple Silicon nodes), **Z-Image-Turbo** (z-image-turbo, text-to-image, ~5 seconds per 1024px image), and **Multilingual E5 Small** (multilingual-e5-small, 384-dimension embeddings, 512-token input, 100 languages). Because the network is powered by independent nodes, the exact catalog changes over time — always query GET /v1/models for the current list rather than hard-coding it; embedding models are listed there with `"type": "embedding"`.

Image generation example:
```bash
curl https://paraloncloud.com/v1/images/generations \
  -H "Authorization: Bearer prlc_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"model": "z-image-turbo", "prompt": "a photo of a graphics card on a black background", "size": "1024x1024"}'
```

Embeddings example:
```bash
curl https://paraloncloud.com/v1/embeddings \
  -H "Authorization: Bearer prlc_your_key_here" \
  -H "Content-Type: application/json" \
  -d '{"model": "multilingual-e5-small", "input": ["query: where is my order", "passage: Your order shipped on Monday."]}'
```

Great for: RAG (retrieval-augmented generation), AI agents, chatbots and conversational AI, semantic search and embeddings, summarization, classification, and any LLM-backed app.

Keywords: inference API, OpenAI-compatible API, serverless LLM inference, LLM API, open model API, free LLM API, free vision API, image generation API, text to image API, embeddings API, text embeddings API, OpenAI-compatible embeddings, decentralized inference, RAG API, AI agent API.

## Credits, Billing & Payments

- One credit balance. The same balance pays for everything on the network — GPU rentals and (later) inference.
- Top up with crypto. Add credits with USDC or USDT on Ethereum or Solana by connecting your wallet. Minimum top-up is $1, credits are added instantly, and they never expire.
- Rentals: billed per minute, pay only for what you use.
- Inference: free during the early network phase; usage is metered in tokens for the future.
- Provider payouts: real USDC on Solana, withdrawn to a linked wallet (no fees), or converted to credits at 1 USDC = 1 credit.

## Referral Program — Earn Up to 30%

Invite anyone to ParalonCloud and earn a share of the platform fee on the rentals they take part in, automatically, for 12 months.

- Earn 15% of the platform fee from every rental your referee participates in.
- It works whichever side they join — whether the person you invite comes to rent GPUs or to provide them.
- Both sides stack: a rental has two parties (renter and provider). If both are your referees, you earn 15% from each side — up to 30% of that rental's platform fee.
- Lasts 12 months from when the referee applies your code. No cap on referrals; referees pay nothing extra.
- Get your code and invite link in Settings.

Keywords: referral program, earn referral rewards, invite and earn crypto, GPU referral.

## Blender Cloud Render (Paralon Render)

- GPU-accelerated Blender rendering — upload .blend files, render in the cloud.
- Cycles and Eevee engines, resolutions up to 4K and beyond, animation batch rendering.
- NVIDIA RTX acceleration; output PNG, JPEG, OpenEXR, MP4; Blender 4.5 support.
- Free during beta. URL: https://render.paraloncloud.com

## Use Cases

- AI model training and LLM fine-tuning
- Machine-learning inference and serving
- RAG systems, AI agents, and chatbots
- Image and video generation (Stable Diffusion, etc.)
- Semantic search and embeddings
- Scientific and high-performance computing
- Blender 3D rendering, animation, architectural visualization, VFX
- Monetizing idle GPUs for USDC income

## Accounts & Sign-In

- Sign in with Google or a one-time email code (passwordless OTP) — no password required.
- Set a public display name for the leaderboard and connect Discord in Settings.

## Documentation

Full documentation lives at [https://paraloncloud.com/docs](https://paraloncloud.com/docs).

- [Introduction: rent GPUs, run inference, or provide compute](https://paraloncloud.com/docs)
- [Inference API overview](https://paraloncloud.com/docs/inference-api)
- [Create API keys and authenticate](https://paraloncloud.com/docs/inference-api/authentication)
- [Chat completions, parameters, streaming](https://paraloncloud.com/docs/inference-api/chat-completions)
- [List and choose models](https://paraloncloud.com/docs/inference-api/models)
- [Embeddings: /v1/embeddings request, response, models and pricing](https://paraloncloud.com/docs/inference-api/embeddings)
- [Responses API: /v1/responses request items, tools, streaming events, limits](https://paraloncloud.com/docs/inference-api/responses)
- [Batch API: JSONL files, /v1/batches lifecycle, output files, 50% pricing](https://paraloncloud.com/docs/inference-api/batch)
- [Rental API overview: rent GPUs programmatically over HTTP](https://paraloncloud.com/docs/rental-api)
- [Rental API authentication and the rental scope](https://paraloncloud.com/docs/rental-api/authentication)
- [List available GPUs](https://paraloncloud.com/docs/rental-api/browse-gpus)
- [Create, poll, list, and destroy rentals](https://paraloncloud.com/docs/rental-api/manage-rentals)
- [MCP server: rent GPUs from Claude Code / Claude Desktop / Cursor](https://paraloncloud.com/docs/rental-api/mcp)
- [How to rent a GPU in the browser, step by step](https://paraloncloud.com/docs/rent-a-gpu)
- [Provider hardware/software requirements](https://paraloncloud.com/docs/prerequisites)
- [Connect your GPU with one Docker command](https://paraloncloud.com/docs/add-a-node)
- [Provider rewards: USDC earnings + incentivized testnet points](https://paraloncloud.com/docs/rewards)
- [Withdraw USDC to a Solana wallet, or convert to credits](https://paraloncloud.com/docs/payouts)
- [Display name](https://paraloncloud.com/docs/account-profile)
- [Referral program: earn up to 30%](https://paraloncloud.com/docs/referrals)
- [Credits, top-up, and billing](https://paraloncloud.com/docs/billing)

## Key Pages

GPU & Compute:
- [Browse and rent available NVIDIA GPUs](https://paraloncloud.com/rent)
- [GPU rental pricing](https://paraloncloud.com/pricing)
- [API Console: create keys, enable the rental scope, view usage and live rentals](https://paraloncloud.com/console)
- [Manage your nodes and sessions, provide compute](https://paraloncloud.com/dashboard)
- [Your active rentals](https://paraloncloud.com/rentals)
- [Rental API endpoint](https://paraloncloud.com/api/v1): rent GPUs programmatically — see /docs/rental-api
- [Top up credits with crypto (USDC/USDT)](https://paraloncloud.com/add-credits)
- [Provider withdrawals and earnings conversion](https://paraloncloud.com/payouts)

AI & Inference (the inference site is https://ai.paraloncloud.com; the API endpoint is https://paraloncloud.com/v1):
- [AI inference product site](https://ai.paraloncloud.com)
- [API console: create/manage API keys, view usage](https://ai.paraloncloud.com/console)
- [Interactive chat to test models, no code](https://ai.paraloncloud.com/playground)

Community & Network:
- [Top providers ranking (incentivized testnet)](https://paraloncloud.com/leaderboard)
- [Live map of global nodes](https://paraloncloud.com/network)
- [Terms of Service and Acceptable Use](https://paraloncloud.com/terms)
## Technology

- Decentralized provider network with real-time node monitoring
- Secure container isolation per rental
- Outbound secure tunnels — no port forwarding or public IP needed for providers
- WebSocket connections for live updates
- Passwordless email OTP and Google sign-in

## Links

- [Website](https://paraloncloud.com)
- [Documentation](https://paraloncloud.com/docs)
- [Rent GPUs](https://paraloncloud.com/rent)
- [GPU Pricing](https://paraloncloud.com/pricing)
- [Provide Compute / Dashboard](https://paraloncloud.com/dashboard)
- [AI Inference Site](https://ai.paraloncloud.com)
- [AI Inference API endpoint](https://paraloncloud.com/v1)
- [AI Console](https://ai.paraloncloud.com/console)
- [AI Playground](https://ai.paraloncloud.com/playground)
- [Leaderboard](https://paraloncloud.com/leaderboard)
- [Network Map](https://paraloncloud.com/network)
- [Blender Render](https://render.paraloncloud.com)
- [Discord](https://discord.gg/VHn9KevdTK)