P¬P Studio
One API key. Zero retention.
Studio is where you get your key, try every model, read the receipts and pay for what you call. It is also where you design, train and deploy models of your own.
What's inside
Everything a call needs, in one account.
API keys
One pnotp_sk_ key reaches the whole platform. Studio shows it once and keeps only its hash.
Playground
Try any model on the list and read the receipt for each answer. Nothing from the Playground is saved.
Models and pricing
Every model with its price per million tokens: the same list the API serves.
Billing
Your balance, automatic top-ups and every transaction, in one place.
Your own models
Design a model on a visual board, train it on CPU, export it as a PyTorch project and deploy it.
Toby
Answers questions about the platform inside Studio, and from code with pnotp.help.toby.
Calling a model
Keep your code. Change the base URL.
from openai import OpenAI client = OpenAI(base_url="https://api.pnotp.ai/v1", api_key="pnotp_sk_...")reply = client.chat.completions.create( model="claude-sonnet-5-5", messages=[{"role": "user", "content": "Review this clause."}],)The API is OpenAI-compatible. Point an OpenAI client at https://api.pnotp.ai/v1, use a PnotP key, and pick a model from the tracklist. The Python SDK (pip install pnotp) and the R package (install.packages("pnotp")) make the same calls.
PnotP never stores your prompts. Every call returns a receipt signed with Ed25519 that names the provider that served it.
Prompt caching
Optional and temporary. It is on by default where a model supports it, scoped to your account and held in memory only: Claude's cache expires after 5 minutes without use, and Google's is set to volatile memory. Turn it off for any call with "prompt_cache": false.
Your own models
Design it, train it, ship it.
Lay out a neural network on the board, train it on CPU, and export a PyTorch project with its source, configs and a model card. Then serve it behind your key.
Board model
Train it on the board, export it, then serve its checkpoint.
Your checkpoint
Upload a TorchScript .pt file and serve it behind your key.
Open-source models
Serve a vision model from the open-source catalog.
Agent
Run a hosted agent with a budget set for each run.
Mark a deployment compliant and its endpoint keeps no request payload, redacts errors and returns a compliance receipt with every prediction.
Pricing
Credits, by the call.
A new account starts with no credits. Model calls bill per token at each model's published price, and agent runs draw from your credits. Training on CPU, exporting and deploying don't charge, but they need credits in the account.
Run
$19 / month
A monthly credit allowance that resets each cycle. Run Plus, $99 a month, carries 5× the allowance. Run Max, $199 a month, carries 10×.
Subscribe in Studio