paratropeBeta

Your own AI thatplays by your rules.

An AI assistant with your notes as context, and the power of a GPU only when it's worth it.

Summarize what changed in the lease since the last version.

Starting a GPU Loading the model Ready

Three changes: the term goes from three to five years (clause 3), notice drops to 60 days (12.2), and a new clause 14 caps liability.

Example

A rented GPU bills
whether you use it or not.

GPUs for AI are rented by the hour, and the meter runs from the moment one is turned on until someone turns it off. Starting one takes minutes, so people leave it on between tasks, or all day, and most of what they pay for is waiting. paratrope turns the GPU on when a task needs it and terminates it twenty minutes after you stop. You pay for the work, not for the wait.

Example: one working day, three tasks, on an A40 GPU at $0.60 an hour

GPU left on

12 h · $7.20

With paratrope

2 h 45 min · $1.65

You save$5.55

that day, 77% less. Over a month of 22 working days, $122.10.

Working
Starting up, and 20 minutes before it's terminated
Billed, unused

You choose the GPU
and the model.

A conversation can stay on the free model on our server, or use a GPU for the task at hand. paratrope suggests the cheapest one that fits; if you'd rather, you pick it yourself from the full catalog, along with the model to run on it.

01 · For each task

Say what it's for and see the price.

Talk in depth, code, or work with images: for each purpose it shows the model, the GPU, the price per hour and how long it takes to be ready. Nothing turns on until you confirm.

What do you want to talk with?✕

PrivatefreeRuns on our server. A small model, for everyday things.Qwen3-1.7B · on our server

Talk in depth~$0.60/hLonger, better-reasoned answers, much faster.Qwen3-14B · A40 GPU · ready in ~3 min

Code~$0.60/hClones your repos, edits files and runs the tests.Devstral Small 2 24B · A40 GPU · ~5 min

With images~$0.60/hUnderstands photos and screenshots.Qwen3.5-35B-A3B · A40 GPU · ~6 min

DoneTurn on and use · ~$0.60/h

GPUs bill while they're on and terminate themselves after 20 min without use. Choose the GPU by hand

02 · By hand

Any GPU and any model that fits.

Browse every GPU with its price, memory and stock, set a budget, and choose the model. It tells you which models fit in that GPU's memory and which don't. You can switch models without turning the GPU off.

Up to $1.00/h

RTX A5000 $0.27/h 24 GB of memory · High stock

L4 $0.59/h 24 GB of memory · Medium stock

A40 $0.60/h 48 GB of memory · High stock

RTX 4090 $0.89/h 24 GB of memory · Low stock

L40S $1.09/h 48 GB of memory · Medium stock

A100 80 GB $1.79/h 80 GB of memory · Low stock

A40 48 GB

Total per hour$0.61

Model

Qwen3-14BRecommended

Mistral-Small-24BRecommended

Qwen3.5-35B-A3BFast

Llama-3.3-70BDoesn't fit

Turn on with Qwen3-14B · $0.61/h

03 · While it's on

You always know what it's costing.

How long it's been on, how much it has cost so far, and how long until it's terminated if you stop using it. Each session stays in the month's history.

Terminated, not paused

A paused machine still bills for its disk. A terminated one bills nothing. And if a machine starts badly, it's terminated right away: you don't pay for a broken one.

Everyday work, on our server

A small model is always on, on our own server. The GPU and web searches are used only when a task calls for them, and you always see when.

Your notes, as context

Notes and documents you choose become context for its answers. You decide what it may do without asking you.

paratrope is in beta and invite-only.

Leave your email and we'll send you an invitation as spots open up.