Build your application once. Hangar runs it where it has to run, within the limits you set, and returns a record anyone can verify.
hangar run --cpu 4 --image doc-eval:2026-10-22 --stop-after 2h \ -- python evaluate.py --set held-out CPU job HG-C9P4WD: 4 vCPU, 8 GiB, AWS Sydney. A$0.48 an hour, A$5.00 minimum. Image doc-eval:2026-10-22 is sha256:1e7b30c4a9d5. The job uses the image with this digest. Starting. Running ✓ on a fresh instance since 15:12:09 AEDT. It stops by 17:12:09. Follow it with hangar logs HG-C9P4WD --follow
The pool
Run your application on Australian capacity. Choose where each job happens.
One contract. Your release runs the same on hosted capacity today and inside a customer's environment from 2027. Your organisation chooses where each job happens.
| Aspect | Hosted in Sydney | Your own environment, from 2027 |
|---|---|---|
| Capacity | NVIDIA H100 and CPU capacity in AWS Sydney (ap-southeast-2). | Your organisation's own AWS account in ap-southeast-2. |
| Data | Stays in Australia. Your outputs come back to you. | Stays in your environment, under your organisation's access control. |
| Platform | TALOS Defence operates and supports Hangar. | TALOS Defence licenses and supports Hangar. Your people, TALOS Defence or a partner operate it. |
The limits
Your organisation sets the limits of each job: the image, the identities, the data, the destinations, the compute and the keys.
TALOS Warrant, the part of Hangar that enforces limits, applies the limits before the job starts and keeps the job within them.
| Who | Controls |
|---|---|
| Your organisation | Where each job happens, which images may start, the limits of each job, and the data each job may read. |
| The vendor | The vendor's application and each image of the application. A new image starts only when your policy allows the image. |
| Hangar | The execution platform, hosted capacity in Sydney and the signed certificate. TALOS Defence operates it. |
The certificate
Every job has a certificate. TALOS Warrant keeps the job within its limits, and TALOS Defence signs the certificate when the job ends.
The certificate states what ran, on what, when, and every move of its data. Your assessor draws the conclusion.
TALOS Defence signs the certificate of every hosted job. Anyone can verify it offline with the published verifier and keys, for at least seven years from issue.
Job certificate · issued by TALOS Defence Pty Ltd
HG-7K3M2Q
No project per customer. Deployment, update and support are the platform's work, not yours.
Book hours on an NVIDIA H100 in Sydney. The job starts on the minute and ends in a signed certificate.
Start a CPU job on demand from the command line or the API, with the same certificate.
The same release runs inside the customer's own environment from 2027, with the same limits and the same signed certificate.
Schedule
CPU capacity is on demand. Nights of H100 capacity appear here when they are scheduled.
CPU jobs start on demand, with the same signed certificate.
hangar run --cpu 2 -- python job.pyBook a job from one hour up to 21 hours on a night, or book the whole night.
| Option | What it covers | Price | Action |
|---|---|---|---|
| H100, by the hourFrom one hour | Your release runs on one NVIDIA H100 in AWS Sydney. Book whole hours ahead, from one hour, or start a job now and pay by the started minute while the block's instance is free. | A$34.95per hour, including GST | Start a job, H100, by the hour |
| Trainium2, by the minuteStart now | Your release runs on one AWS Trainium2 in AWS Melbourne, on a fresh instance for your organisation alone. Billed by the started minute from prepaid credit. | A$9.99per hour, including GST | Start a job on Trainium2, Trainium2, by the minute |
| The whole night, your jobs only | Only your jobs use the H100 that night: 24 hours, 22:30 to 22:30 Melbourne time. | A$595.00per night, including GST | Book the night, The whole night, your jobs only |
| CPU, by the vCPU-hourOn demand | Your release runs on a fresh CPU instance in AWS Sydney, for agent and data work. The job starts when you start it. A$5.00 minimum. | A$0.12per vCPU-hour, including GST | Start a job on CPU, CPU, by the vCPU-hour |
Prices in Australian dollars. Every price includes GST. You pay from prepaid credit, and paid credit never expires.
Pricing detailsYour application, as a container or a script with its data. Hangar runs the application within the limits you set, on an NVIDIA H100 or a CPU instance in AWS Sydney.
The certificate is the signed record that a job ends with. It states what ran, on what, when, and every move of the job's data. TALOS Defence signs it, and anyone can verify it offline with the published verifier and keys.
When you confirm the hold. A hold lasts 10 minutes. Your credit is charged only when you confirm.
From one hour up to 21 bookable hours on a night. The whole night gives your jobs all 24 hours, 22:30 to 22:30 Melbourne time.
Cancel a booking more than 48 hours before it starts for full credit back. Inside 48 hours its hours are offered for booking again, and your credit returns for the hours that are booked.
Your work runs in AWS Sydney (ap-southeast-2). Data and storage stay in Australia. Monitoring reads account and instance metadata only.
Yes. Keep data that Victorian law requires to stay in Victoria out of Hangar until Hangar offers a Melbourne region. Read Victorian data residency.