AI orchestration
Work that runs itself.
Turn one-off notebook work into pipelines that run on a schedule or on demand: prepare the data, train, check and publish, in order, every time.
Who it's for
Repeatable training and data jobs.
Analytics teams
Refresh models and reports every night without anyone pressing a button.
ML engineers
Spread big jobs across many GPUs with Ray, without managing a cluster.
Operations
Know that every step ran, in order, with a record of what happened.
What teams use it for
- Retrain a fraud model every night on the day's transactions
- Clean and load new data every morning before work starts
- Run a large evaluation across several GPUs at once
How it works
Three steps.
- 01
Describe the steps
As a Kubeflow pipeline, an Airflow schedule or a Ray job.
- 02
Set when it runs
On a schedule, when new data arrives, or by hand.
- 03
Watch it run
Each step's status, logs and results in one view.
What you get
AI orchestration, in detail.
Kubeflow pipelines
Chain data preparation, training and checks into one repeatable run.
Airflow scheduling
Run workflows on a timetable, with retries when a step fails.
Ray for scale
Spread work across several GPUs from ordinary Python code.
From notebook to pipeline
Kale turns a working notebook into a pipeline without rewriting it.
Only pay while it runs
GPUs are used for the steps that need them and billed by the second.
Under the hood
The key numbers.
- Billing
- Per second
- Minimum per run
- 60 seconds
- Location
- Kathmandu Valley, Nepal
- Billing
- Per second
- Minimum per run
- 60 seconds
- Location
- Kathmandu Valley, Nepal
- GPU
- NVIDIA H200 NVL
- GPU memory
- 141 GB
- Memory bandwidth
- 4.8 TB/s
- Currency
- Nepalese rupees (NPR)
- Ways to pay
- Digital payment, bank transfer or invoice
Billing
GPU time used, per second, in NPR
Commitment: none. Rates are being finalised; every request gets a firm quote in rupees.
How billing worksQuestions
About ai orchestration.
Do we have to use all three tools?
No. Use whichever fits: Kubeflow for ML pipelines, Airflow for schedules, Ray for distributed work.
What happens if a step fails?
The run stops at that step, keeps the logs, and can retry or resume from there.
Start with ai orchestration.
Tell us what you want to run. It takes about a minute.