nebula@atheron:~$
// Machine learning on Constellation
Big jobs, many GPUs.
Open models. Checked results.
Atheron Nebula runs your batch inference and fine-tuning on the ATON Constellation network: GPUs shared by members, plus Atheron's own house nodes. Every result is verified, and you pay from prepaid credit.
// How it works
Send work, get results
You make API calls. We find the GPUs, check the answers and give you one clean result.
- 1
Set up your team
Create your team in the developer portal, pass a quick business check and add a card.
- 2
Send a job
Submit a batch or a fine-tuning job with your API key.
- 3
We run and check it
Members' GPUs take the work first, Atheron house nodes fill in when not enough members are online, and every part is verified.
- 4
Get your results
Download one result file, or have a webhook tell you it is ready.
// Services
What Nebula does
Two services to start, with more on the way.
Batch inference
Send a large batch of text, audio or images and get embeddings, labels, transcripts or captions back.
Fine-tuning
Teach an open model your task with LoRA fine-tuning, on your private data, and download the adapter.
Coming soon
Hosted open-model APIs, image generation and video generation are next. Join the waitlist.
For members
Put your GPU to work through ATON Constellation and choose points, cash or USDC for each job.
// Why Nebula
Built to be trusted
Running work on other people's machines only works if the rules are strict. Ours are.
Verified answers
Each shard runs on at least 2 machines owned by different people, and an Atheron house node settles any difference.
Private by default
Your data runs only on Atheron house nodes or AWS GPUs unless you mark it shareable.
A vetted catalogue
Only open models we have checked and signed run on the network. Nobody can swap a model under you.
Prepaid, never a surprise
A plan brings credit each month and top-ups add more. Every job reserves its estimate plus a quarter before it runs, and you pay what it costs, never more.
// Pricing
Pay for what you run
Each service has one simple unit, like a million tokens or an audio hour. Prices are announced at opening.
- Embeddings, small and base modelsper million tokens
- Embeddings, large modelsper million tokens
- Classificationper 1,000 items
- Transcriptionper audio hour
- Captionsper 1,000 images
- Fine-tuning (LoRA)per GPU hour
// For members
Have a GPU? Put it to work.
Run Nebula jobs through the ATON Constellation app while your computer is idle, and choose points, cash or USDC for each job. Cash and USDC open after verification.
Be first in line
Leave your address and we will write when Nebula opens. One confirmation email first, then only news about Nebula.