Standard APIs. Copy, paste, ship.

Aurora speaks the S3 API, the OpenAI API, and SSH. If your tools work with those, they already work with Aurora.

Developer demos

Recipes for every service.

Six common tasks, from first inference call to giving agents their own office.

chat.py

01import os02from openai import OpenAI0304client = OpenAI(05    base_url="https://ai.aur.lu/v1",06    api_key=os.environ["AURORA_API_TOKEN"],07)0809reply = client.chat.completions.create(10    model="deepseek/deepseek-v4.1-flash",11    messages=[{"role": "user", "content": "Summarize this contract."}],12)13print(reply.choices[0].message.content)

Point the official SDK at Aurora. Nothing else changes.

Open standards

Familiar interfaces. Infrastructure you control.

Standards-based APIs across the stack mean your code moves without rework—onto Aurora, or away from it.

inference.py

01import os02from openai import OpenAI0304# Point any OpenAI-compatible client at Aurora05client = OpenAI(06    base_url="https://ai.aur.lu/v1",07    api_key=os.environ["AURORA_API_TOKEN"],08)0910reply = client.chat.completions.create(11    model="deepseek/deepseek-v4.1-flash",12    messages=[{"role": "user", "content": "Hello!"}],13)

Use your OpenAI SDK with open-weight models.

Keep your SDKs and prompts. Change the base URL and key, test against your own evals, and pick the right model per call.

zsh · s3

S3 API

$ aws s3 cp ./checkpoints s3://training-runs/ \    --recursive --endpoint-url https://a-s3.aur.luupload: checkpoints/step-24000.safetensorsupload: checkpoints/step-25000.safetensorsupload: checkpoints/optimizer.pt$ aws s3 ls s3://training-runs/ --summarize \    --endpoint-url https://a-s3.aur.luTotal Objects: 3   Total Size: 412.6 GiB

Point your existing S3 tools at Aurora storage.

The AWS CLI, boto3, and rclone talk to Aurora as they do today for common operations, and S3 API requests aren’t charged.

inference / request log

metadata only

Time	Model	In	Out
14:02:11	deepseek-v4.1-flash	1,204	312
14:02:09	kimi-k3	8,930	1,044
14:01:57	glm-5.2	2,410	688
14:01:40	deepseek-v4-pro	15,022	2,301

Prompt: not stored
Output: not stored
Training: never

Private inference that never stores your prompts.

Prompts and outputs are processed transiently in your region. Never stored, never used for training, never proxied to a third-party API.

console.aurora / resources

All resources+ New

Name	Type	Status
deepseek-v4-pro	Inference	Active
training-runs	Bucket	412 GiB
acme-prod	VPC	4 machines
scratch-01	Volume	200 GiB
etl-worker-02	VM	Running

Manage storage, compute and access in one console.

Buckets, credentials, VMs, networks, volumes, and an event log, so your team can see what changed and who changed it.

Resources

Everything else you need.

DocumentationGuides, API reference, regions, and limits for every service.Aurora portalCreate keys, buckets, machines, and networks—and watch the event log.xbin on GitHubThe open-source agent workspace. MIT / Apache-2.0, single Go binary.

Open the portal. See it for yourself.

Create an account, make a bucket, and send your first inference call in minutes. No sales call, no commitment.

Start free Connect with us

30-day free trial · no payment method required