LLM Hangar / Private LLM hosting
Private LLM hosting in your own AWS, Nebius or RunPod account
Private LLM hosting usually means one of two things: a vendor runs the model on their GPUs and promises not to look, or you run it yourself and own the operations. LLM Hangar is the third option. It deploys an open model onto a GPU instance inside your own cloud account, in the region you choose, and hands you a private endpoint. The infrastructure is yours; the platform only acts in your account, through access you can revoke, and writes down everything it did.
What "private" means here
- Prompts never transit LLM Hangar. Requests go from your client straight to the endpoint on your instance, over HTTPS. There is no proxy in the middle to store, log or train on your content.
- Sub-processors for prompt data: none. The only party that handles your prompts is your own cloud provider, under the agreement you already have with them. See the privacy policy for what the platform itself holds.
- The model runs on your machine. Weights are downloaded onto the instance disk in your account. No shared tenancy, no queue behind other customers, no third party holding your model.
- One endpoint key, stored as a hash. It is shown once. Rotate it, or delete the deployment, and it stops working immediately.
The access LLM Hangar holds in your account, and how it is scoped, is described on the security page.
What you get
- An OpenAI-compatible endpoint with a valid certificate. Existing Python, JavaScript, LangChain, Vercel AI SDK and editor configurations work by changing the base URL (snippets).
- A hard budget cap on every deployment. You can raise it; you cannot disable it. At the cap the deployment is destroyed and you get an email naming it and the amount.
- A self-destruct timer, so a forgotten GPU cannot run all weekend. Trial deployments always carry one.
- Verified teardown. After every destroy we check with your provider that the resources are actually gone, and keep checking until that is confirmed.
- An audit log of every action taken in your account, timestamped, including the plan recorded before anything is created. Exportable as JSON.
- An EU-only checkbox that pins every resource of a deployment to EU member-state regions and keeps it there.
What it costs
Two bills. LLM Hangar is a flat $39 a month on the Lab plan (€34 in EUR), with a 24-hour trial that requires a card. Your cloud provider bills you directly for the GPU hours, at your own provider's rates, with your own credits and reserved capacity if you have them. We never see that invoice.
Two reference points for the GPU side, both cited rather than estimated:
- On AWS, a g6e.xlarge (one NVIDIA L40S, 48 GB) was listed by Spare Cores on 2026-08-22 at $1.974 per hour on-demand and $0.604 spot in Stockholm, and $2.327 / $1.747 in Frankfurt. That shape runs models up to roughly 27B parameters at 4-bit.
- On RunPod, the 2x H200 shape that runs DeepSeek V4 Flash cost $7 to $14 per hour at our own boots in August 2026, the rate fixed at pod creation. The full numbers are in the DeepSeek V4 Flash guide.
The wizard shows the exact hourly and monthly estimate for the shape and region you pick before you confirm anything, and a cost calculator lets you compare that against a per-token API at your own volume.
Who it is for
- Teams that are not allowed to send data to third parties. Client data, source code, medical or legal records: the prompt stays on a machine in an account you control.
- EU teams who need every resource in EU regions without adding a new vendor to their sub-processor list. See EU-hosted LLM API providers compared.
- Solo builders and small teams who want a private endpoint for an evening's prototype and do not want to write Terraform, build an AMI or configure an inference engine to get it.
Honest limits
- The Lab plan runs one GPU shape per deployment, as listed on the plan card. The largest catalog models need multi-GPU shapes; the per-model guides state which.
- LLM Hangar does not currently hold a SOC 2 or ISO 27001 certification. The controls are described so you can evaluate them directly; the audit log and your own console let you verify them.
- Budget caps and timers are enforced against estimates from catalog prices. Your provider's meter, not our estimate, is the invoice. Caps bound the window in which spend can accumulate; they are not a guarantee of the exact amount.
Frequently asked questions
Is this the Private LLM app?
No. Private LLM is an iOS and Mac app that runs small models on your device. LLM Hangar is hosting infrastructure: it deploys open models onto GPU instances inside your own AWS, Nebius or RunPod account and gives your applications a private, key-authenticated, OpenAI-compatible endpoint.
Do my prompts ever reach LLM Hangar?
No. Requests travel directly between your client and the endpoint on your instance. LLM Hangar does not proxy, store, log or train on prompt or response content, and there are no sub-processors for prompt data. What LLM Hangar holds is your account data and an audit log of the infrastructure actions it performed for you.
Can I keep the infrastructure if I cancel?
Yes. The instance, disk, network and endpoint are resources in your own account, tagged and owned by you. They stay after you cancel; you can manage them in your own console. Access for LLM Hangar is granted through a role or key you can revoke at any time.
Which providers and regions?
AWS, Nebius and RunPod today. Regions come from your linked provider. The EU-only option pins every resource of a deployment to EU member-state regions, for example eu-central-1, eu-west-1, eu-west-3, eu-north-1, eu-south-1 and eu-south-2 on AWS.