Inference assessment
Model selection, GPU sizing, hosting choice and a monthly cost estimate, delivered as a written assessment and deployment plan.
$249 · one-time
DomainOS Managed Inference
Your own private model API, in your cloud, run for you.
We deploy an appropriately licensed open-weight model in your own cloud account and deliver a secured, tested API endpoint. Then we manage it.
Assessment $249 · deployment from $499 · management from $199/month · GPU billed to you
Want to talk first? Book a 30-minute consultation
Every request, inside your account
Your applicationHTTPS + your API key
Restricted gatewayTwo routes allowed · everything else closed
vLLM serving engineLoopback only · never publicLicensed open-weight modelPinned revision
GPU billed to you · DomainOS access revocable
Short demonstration · 1:22
See DomainOS Managed Inference in action
Watch how we deploy an open-weight AI model to cloud GPU infrastructure, expose it through an authenticated API, and connect it to an application.
Prefer to talk first? Book a 30-minute consultationTutorial chapters and notes →
- ModelOpen-weight Qwen deployment
- ComputeNVIDIA cloud GPU infrastructure
- ServingvLLM inference serving
- AccessAuthenticated HTTPS API
- Integration
Python application integration - EvidenceInference testing and cost visibility
Your cloud
RunpodAWSGoogle CloudAzureLambdaServing
vLLM
Hugging FaceModels
QwenMeta LlamaMistral AIGoogle GemmaFrom the Engineering Lab · Full tutorial · 39:47
The full build, end to end, in one tutorial
A complete hands-on walkthrough of running Qwen locally, deploying it to a cloud GPU with Modal and vLLM, exposing an authenticated API, testing inference, and inspecting actual compute costs.
What you watch us build
- 01Choose and run an open-weight model (Qwen) locally
- 02Deploy it to a cloud GPU with Modal and vLLM
- 03Put it behind an authenticated, OpenAI-compatible API
- 04Test inference and read the real compute bill
This is the work we do for clients, in your own cloud account: deployed, tested, documented and handed over, then kept running.
Have us deploy yoursFree learning series · Assess → Deploy → Operate
- 01 · AssessInference AssessmentModel selection, GPU sizing, hosting decisions, benchmarks and cost estimation.Upcoming
- 02 · DeployDeploy your own AI model to the cloudQwen on a cloud GPU with Modal and vLLM, behind an authenticated API.Published
- 03 · OperateInference OperationsMonitoring, troubleshooting, updates, performance optimisation and cost management.Upcoming
Our services
Three services: assess, deploy, manage.
Start with whichever you need. Each one is tracked in your DomainOS workspace, where you follow the work, review what we deliver and export it.
Managed inference deployment
We deploy your model, configure the API and authentication, test it, document it and hand it over.
From $499 · one-time
Ongoing management
Monitoring, updates, troubleshooting, and regular performance and cost reviews.
From $199 / month
Who it’s for
Teams with a real application and a deployment problem.
One model, one endpoint, one integration example, done properly. If you need fine-tuning, regulated data handling or a 24/7 guarantee, we will say so on the call and scope it separately.
AI product teams
You have a prototype on an open-weight model and no stable, private endpoint to put behind it.
Engineering teams
You are moving a Qwen or Llama proof of concept into a cloud account your company controls.
Agencies and integrators
You build AI products for clients and want model deployment and upkeep handled for you.
Exact scope
What you get, and what you don’t.
Written into your one-page quote before you pay anything.
Included
- Discovery call and written workload and capacity assumptions
- Licence review of one supported open-weight model
- One private deployment in your cloud account
- One authenticated HTTPS chat-completions endpoint
- Sample Python and curl integration
- Smoke tests and an evaluation on approved examples
- Scheduled checks, version records and spend notes
- Runbook, export and shutdown instructions, handover meeting
- Up to 2 hours of weekday best-effort support each period
Not included (quoted separately)
- Fine-tuning, adapters, retrieval (RAG), tools or agents
- Ingesting your private company documents
- 24/7 support, an uptime guarantee or high availability
- Compliance accreditation or regulated-data handling
- Autoscaling, multiple regions or multiple models
- Building your web or mobile application
How it works
From application to a managed endpoint in seven steps.
Everything happens in your DomainOS workspace: book the call, review the quote, pay, follow the deployment and sign it off. No unpaid work starts, and no GPU is bought on your behalf.
- 01
Apply
Five short questions about your application, workload, cloud and timing. About three minutes.
- 02
Discovery call
Cal.comThirty minutes, booked straight from DomainOS. We confirm fit, model and GPU budget.
- 03
One-page quote
Model, GPU, region, a monthly compute forecast with a spend cap, and the exact scope.
- 04
Pay
ContraAccept the quote and pay the setup fee on Contra ($499 for a basic deployment). Upload the receipt in DomainOS; we confirm it.
- 05
Deploy and test
We deploy in your account and run every acceptance check, visible to you as it happens.
- 06
Accept
Call the endpoint from your own environment. Managed service starts when you sign off.
- 07
Run it
Usage, billing, support, model changes and cancellation, all from your DomainOS workspace.
Delivered means tested
We don’t call it done until it passes.
The model server is never exposed directly: some serving routes are not covered by its own key check, so a separate gateway allows only the two routes you need. You watch each acceptance check pass in DomainOS, then call the endpoint yourself before signing off.
- A request with your key returns a response from the agreed model
- Missing or wrong keys are refused on generation and model routes
- Admin, debug and other serving routes are closed to the internet
- Only the gateway port is public; the model server is not
- TLS verified; no secret in code or handover documents
- Endpoint use and your cloud spend are visible and recorded
- Model, licence, revision and configuration are documented
- Export, shutdown and termination steps are demonstrated
Who owns what
Your assets stay yours, and you can leave with them.
- 01Your dataYours. DomainOS never claims it.
- 02Cloud account and computeYours, billed to you by your provider.
- 03Configuration and evaluation examplesYours, exported on request and on exit.
- 04Base model weightsUsed under the publisher’s licence; nobody gains exclusive ownership.
- 05DomainOS tools and platformRemain DomainOS’s; you receive the service.
- 06Operating accessTime-limited, least-privilege, revocable by you at any time.
Inside DomainOS
Run the whole service from one workspace.
Usage, spend, checks, invoices and support requests in one place, recorded daily from your provider’s console and our scheduled checks.
Service status
Where your deployment stands and the one thing to do next.
Deployment
Endpoint, model and revision, licence, serving stack and the acceptance checklist.
Usage and allowances
Requests, tokens, latency, scheduled checks and cloud spend against your approved cap.
Billing
Invoices, payment details and receipts, renewals and what you have paid through.
Models
The supported models with licences and hardware; ask to switch with an estimate first.
Plan and support
Requests with time logged against your allowance, add-ons, and cancel any time.
Code samples for
curlPython
Node.js
OpenAI SDKSupported models (licence checked for the exact revision)
- Qwen3 8B Apache-2.0
- Qwen2.5 7B Instruct Apache-2.0
- Llama 3.1 8B Instruct Llama 3.1 Community
- Mistral 7B Instruct Apache-2.0
- Gemma 2 9B Gemma Terms
- Mistral Small 24B Apache-2.0
- Qwen2.5 32B Instruct Apache-2.0
FAQ
Straight answers.
Why not use Hugging Face, Baseten or another hosted provider?
If self-service is enough, use it. We sell configuration, a private environment in your own account, testing, handover and ongoing management, for teams that need the endpoint in infrastructure they control.
Who pays for the GPU?
You do, directly to your cloud provider. DomainOS never resells compute in this plan. Your quote includes a monthly forecast and a spend cap; we stop and ask before going over it.
What does the monthly plan cover?
Essential ($199 a month) covers health checks, troubleshooting, a usage and cost review, routine configuration help and up to two engineering hours. Growth and Advanced add more frequent reviews, updates, optimisation and more hours.
Do we own the model?
You own your data, configuration and evaluation examples, and they are exported to you on request. Base model weights stay under their publisher’s licence; we document exact rights and test the exit path.
Do you fine-tune?
Not in this plan. We test the untuned model on your examples first; tuning is quoted separately only if it falls short of an agreed evaluation.
Can we cancel?
Any time from your DomainOS workspace. The service runs to the end of the period you have paid for, then we hand over the export and shutdown steps.
How do we pay?
Through Contra. After you accept your quote, DomainOS links you to the setup payment; the monthly service is a Contra subscription you start at handover. Upload the receipt in the app and we confirm it, usually within one business day.
Is this suitable for sensitive or regulated data?
Not under the Basic deployment, which uses synthetic or approved non-sensitive examples. Sensitive workloads are an Enterprise / Dedicated Infrastructure project with a separately reviewed security, privacy and legal scope.
Get started
Get your private endpoint into production.
Apply in three minutes, book a 30-minute call, and get a one-page quote with your GPU forecast. Nothing is charged until you accept it.
Building something bigger, like a Domain Intelligence Layer? See IntelligenceOS