Documentation
Introduction
PETENTIA is a decentralized AI compute platform. Rent GPUs, deploy AI models, call them through one API, and verify every compute minute you pay for.
What you can do
- GPU Marketplace — discover and rent NVIDIA and AMD GPUs by the hour.
- AI Playground — generate images, video, text and voice with hosted models.
- Serverless AI — deploy Hugging Face models and custom containers.
- Developer API — use PETENTIA from your own applications.
- GPU Instances — manage rented machines over SSH.
- Billing — track compute usage and spending with prepaid credits.
- On-chain Receipts — verify billed usage with Merkle proofs.
- PETEN ecosystem — the token that connects compute demand and supply.
Who it is for
PETENTIA is built for ML engineers who need a GPU for an afternoon, startups serving models in production, and builders who want transparent, crypto-native billing without contracts or credit cards.
Supported networks
Overview
Getting started
Go from zero to a running GPU in five steps.
1. Connect your wallet
Open the console and press Connect Wallet. Choose a network and a wallet you have installed. Your public address becomes your account — there is no password and PETENTIA never asks for your seed phrase.
2. Add ecosystem credits
Go to Billing and press Add credits. Pick a supported token, enter an amount and review the conversion before confirming. Credits are prepaid; you only spend what you use.
3. Choose your GPU or AI model
Browse the GPU Marketplace and filter by VRAM, manufacturer or price. For hosted inference, open Serverless Models or the AI Playground.
4. Run your workload
Connect to your instance over SSH, call your serverless endpoint with an API key, or generate directly in the Playground.
5. Track usage and verify receipts
The dashboard shows credits, spend and GPU usage in real time. When a workload stops, its usage is settled into a receipt you can verify under Receipts.
Overview
Preview mode
The console is currently a preview of the full platform. It lets you explore every workflow before production capacity comes online.
In preview mode no GPUs are provisioned, no models are called and no funds move. Wallet sign-in is real; everything else is simulated in your browser.
What is real
- Wallet connection and network switching on supported networks.
- The GPU catalog, specifications and list rates.
- Token prices shown during deposit review (from CoinGecko).
- SHA-256 receipt hashing, Merkle roots and proof verification, computed in your browser.
What is simulated
- Instance provisioning, IP addresses and SSH access.
- Serverless deployments and endpoint requests.
- Credit deposits — no transaction is sent from your wallet.
- Playground outputs — images and video are procedural preview renders, voice uses your browser's speech engine.
- On-chain anchoring of Merkle roots.
Where data is stored
Preview records are saved in your browser's local storage, separately for each connected wallet. Clear them any time from Settings → Reset preview data.
Overview
Wallets & networks
Sign in with the wallet you already use. PETENTIA supports EVM and Solana wallets.
Supported wallets
| Network type | Wallets |
| EVM | MetaMask, Rabby, Coinbase Wallet and any EIP-6963 browser wallet |
| Solana | Phantom, Solflare, Backpack |
Networks
Switching networks
When you connect with an EVM wallet, PETENTIA asks the wallet to switch to the selected network and adds it if needed. If you change network inside your wallet, the console follows automatically.
Security
- Connecting shares only your public address.
- PETENTIA never asks for a seed phrase or private key.
- Always check you are on petentia.xyz before signing anything.
Compute
GPU Marketplace
Rent dedicated GPUs billed per second. Every listing shows VRAM, memory bandwidth, host resources and the hourly rate.
Choosing a GPU
| Workload | Good fit |
| LLM training, 70B+ inference | H200, H100, MI300X |
| Fine-tuning 7B–34B, batch inference | A100 80GB, L40S |
| Image & video generation | L40S, RTX A6000, RTX 5090 |
| Prototyping, small models | RTX 4090, RTX 5090 |
Filters and sorting
Search by name or architecture, filter by minimum VRAM, manufacturer and availability, and sort by hourly price or VRAM.
Renting
- Press Rent GPU on a card.
- Choose the GPU count, an estimated duration and a base image.
- Review the hourly cost and estimated total, then Confirm Rental.
You need enough credits for at least one hour at the selected rate. The duration is only an estimate — you pay for the actual runtime.
Compute
Instances & SSH
Manage the GPU machines you rent from My Instances.
Lifecycle
| Status | Billing | Meaning |
| Provisioning | Yes | The machine is being prepared. |
| Running | Yes | Ready for SSH and workloads. |
| Stopped | No | Paused; usage so far is settled into a receipt. |
| Ended | No | Terminated and disk released. Cannot be restarted. |
Connecting over SSH
Add your SSH public key in Settings before renting so it is attached at launch. Then copy the SSH command from the instance row or detail view:
ssh -i ~/.ssh/id_ed25519 -p 22 root@<instance-ip>
nvidia-smi
ssh -L 8888:localhost:8888 root@<instance-ip> # forward Jupyter
Credits running out
If your balance reaches zero, running instances and models are stopped automatically and their usage is settled.
Compute
AI Playground
A workspace for trying hosted models without writing code.
Modes
- Image — aspect ratio, steps, guidance and seed.
- Video — aspect ratio, duration and frame rate.
- Text — system prompt, temperature and max tokens.
- Voice — voice, speed and pitch.
Output and history
The output panel shows the result, generation status, time and credits used, with a download button. Recent generations are listed in the history panel; click one to reopen it.
Cost
The estimated cost is shown under the Generate button before you run a model. In preview mode generations are not charged.
Compute
Serverless AI
Deploy a model once and call it over HTTPS. Workers scale with traffic and back to zero when idle.
Deployment wizard
- Source — a Hugging Face model ID such as
meta-llama/Llama-3.1-8B-Instruct, or a Docker image with an HTTP port.
- Compute — GPU type, maximum GPU instances, concurrency per instance and idle timeout.
- Environment — variables such as
HF_TOKEN. Values are write-only.
- Review — confirm the endpoint and warm-worker rate, then deploy.
Scaling settings
| Setting | Effect |
| Max GPU instances | Upper bound on parallel workers. |
| Concurrency | Requests one worker handles at the same time before another starts. |
| Idle timeout | How long a worker stays warm after the last request. |
Calling your endpoint
curl -X POST https://petentia.xyz/api/v1/serverless/<your-model> \
-H "Authorization: Bearer $PETENTIA_API_KEY" \
-H "Content-Type: application/json" \
-d '{"input": "Hello"}'
Managing models
Each model card shows its endpoint, GPU, status, requests today and compute usage, with actions to start, stop, edit, delete, view requests and copy the endpoint.
Developers
API reference
A REST API with JSON bodies and Bearer authentication.
Base URL
https://petentia.xyz/api/v1
Authentication
Create a key in API & Keys and send it in the Authorization header. Keys are shown once at creation; store them in an environment variable such as PETENTIA_API_KEY and never commit them to source control.
Authorization: Bearer $PETENTIA_API_KEY
Endpoints
| Method | Path | Description |
| POST | /inference | Run a text, image or audio model |
| GET | /gpus | GPU classes, rates and availability |
| POST | /instances | Rent a GPU instance |
| GET | /instances/{id} | Instance status, runtime and cost |
| POST | /instances/{id}/stop | Stop an instance and settle usage |
| POST | /serverless | Deploy a serverless model |
| GET | /usage | Usage and spend for a date range |
| GET | /receipts/{id} | Receipt with Merkle proof |
Example
Errors
| Status | Meaning |
| 401 | Missing, invalid or revoked API key |
| 402 | Insufficient credits |
| 429 | Daily cap reached or rate limited |
| 503 | No capacity for the requested GPU — retry or choose another |
Spending limits
Every key has a daily spending cap. Requests beyond the cap return 429 until the next UTC day. Disable or revoke a key instantly from the console.
Account
Billing & credits
PETENTIA uses prepaid credits. One credit equals one US dollar of compute.
Adding credits
- Open Billing and press Add credits.
- Select the network and a supported payment token.
- Enter the deposit amount and check the credits you will receive.
- Review the transaction and confirm.
Stablecoins convert one-to-one. Native tokens convert at the market price shown during review.
How usage is metered
- Instances — per second while provisioning or running, at the hourly rate × GPU count.
- Serverless — per second while a worker is warm.
- Playground & API — per generation or per token, at the model's listed price.
Usage history
Billing shows your balance, total spend, a daily spending chart, deposit history and compute usage history.
Account
On-chain receipts
Every settled workload produces a receipt that can be proven to belong to a published hourly Merkle root.
How it works
- When an instance or model stops, its usage is settled into a receipt: resource, duration, rate, credits, time window and wallet.
- The receipt is serialized and hashed with SHA-256 to produce a leaf.
- All leaves in the same hour are combined pairwise into a Merkle tree. The top hash is the hourly root.
- Roots are anchored on-chain. Anyone holding a receipt and its proof can recompute the root and compare it with the anchored value.
Verifying a receipt
Open a receipt in Receipts. Verify proof recomputes the leaf from the receipt data and walks the proof up to the root. Verify on Explorer opens the anchoring transaction once roots are published on-chain.
Statuses
| Status | Meaning |
| Pending | The receipt is in the current hour's open batch. |
| Verified | The batch is closed and its root is final. |
In preview mode roots are computed in your browser and are not anchored on-chain.
Ecosystem
PETEN token
$PETEN is the ecosystem token of PETENTIA.
Contract
Only trust the contract address published on petentia.xyz and the official X account.
Utility
- Paying for GPU time, serverless inference, Playground generations and API usage.
- A shared settlement asset for GPU providers joining the network.
- Participation in ecosystem programs and community proposals.
Learn more
See the PETEN page for the ecosystem roadmap. Token supply, distribution and listings are announced only through official channels.
Ecosystem
FAQ
Is PETENTIA live?
The console runs in preview mode. You can connect a wallet and explore every workflow; hardware provisioning, deposits and on-chain anchoring are enabled as production capacity comes online. See Preview mode.
Do I need an account or email?
No. Your wallet address is your account.
Which GPUs are available?
H200, H100, A100 80GB, AMD Instinct MI300X, L40S, RTX A6000, RTX 5090 and RTX 4090. See the marketplace for specs and rates.
How am I billed?
Per second from prepaid credits. Stopping an instance or model stops billing immediately.
Can I bring my own model?
Yes — deploy any Hugging Face model or a custom Docker container with Serverless AI.
How do I verify what I was charged?
Every settled workload has a receipt with a Merkle proof against an hourly root. See On-chain receipts.
Where is the PETEN contract address?
On the PETEN page. Only trust addresses published on petentia.xyz.
How do I contact the team?
Reach us on X or by email at hello@petentia.xyz.