Atheron Nebula

Batch inference on open models, run on ATON Constellation and Atheron's own machines. Prepaid credit in Canadian dollars: a plan, top-ups, and every job checked before it runs.

Melden Sie sich an, um Ihre eigene Test-Client-ID in den Beispielen zu sehen.

Atheron Nebula runs machine learning jobs for you: embeddings, classification, transcription and image captions, on a signed, checked catalogue of open models. Fine-tuning with LoRA comes next.

How it works

1. Sign up your team on the Nebula page, pass the business check, and add a card. The card stays on file.
2. Choose a plan. Starter, Pro or Scale, each with credit every month. Add top-ups any time.
3. Make a key with the nebula right on one of your apps.
4. Estimate, then submit. Every live job shows its estimate first. When you confirm, the estimate plus a quarter is set aside from your credit, and the job runs.
5. Collect the results when the batch is done.

Test keys (ath_test_) run up to 100 items a batch on Atheron's own machines and are never billed. They need no credit.

Where your jobs run

Nebula runs on ATON Constellation:

• Members' machines first. A batch you mark shareable is cut into shards that members' GPUs run, and each answer is checked against a second one before it counts.
• Atheron's house nodes always fill in. When not enough members are online, Atheron's own nodes run the work, so a job never waits on the network.
• Private jobs stay home. A batch is private unless you mark it shareable, and test batches and fine-tuning data are always private. Private work runs only on Atheron's own house nodes or on AWS, never on a member's machine.

Paid plans set your place in the queue: Scale first, then Pro, then Starter.

Prices

One price everywhere, in Canadian dollars, by unit: a million tokens, a thousand items, an hour of audio, a thousand images, or a GPU hour. Your credit is spent at these prices. Signed in, see the prices, the plans and the top-ups on the Plans page.