> ## Documentation Index
> Fetch the complete documentation index at: https://docs.valkyrieapp.azumo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Subprocessors

> The third-party services Valkyrie relies on to run your workloads, and what data each one handles.

Valkyrie runs on a small set of third-party providers. This page lists every
service that processes your data on our behalf, what it is used for, and what it
can see.

## Subprocessors

| Provider | Purpose | Data it handles | Location |
| - | - | - | - |
| **Amazon Web Services** | Hosts the Valkyrie API and dashboard, stores datasets and fine-tuning outputs (S3), and sends account email | Account data, datasets, fine-tuning outputs, request metadata | United States (us-east-1) |
| **Vast.ai** | Primary GPU provider for fine-tuning jobs and model deployments | Datasets, model weights, and inference prompts and completions while a job or deployment is running | Varies by GPU host |
| **RunPod** | Secondary GPU provider, used when capacity is not available on Vast.ai | Same as Vast.ai | Varies by GPU host |
| **Stripe** | Payment processing for wallet top-ups | Payment details, entered on Stripe's hosted checkout page | United States |
| **Sentry** | Error monitoring for the dashboard | Error reports, which can include page URLs and account identifiers | United States |

## How GPU providers are used

Vast.ai and RunPod are GPU marketplaces. When you start a fine-tuning job or a
deployment, Valkyrie rents a GPU machine from one of them, runs your workload on
it, and releases it when the work is done or the deployment is stopped. Your data
is only present on that machine for the life of the workload.

Because these are marketplaces, the physical machine and its location depend on
which host has capacity at the time. If your data cannot leave a specific region
or must stay on infrastructure you control, talk to your Azumo contact before
uploading it.

## Services that are not subprocessors

* **Hugging Face.** Valkyrie downloads open base models from Hugging Face. Only the
  model name is sent; none of your data is.
* **OpenAI and Anthropic.** Valkyrie's API follows the OpenAI and Anthropic request
  formats so existing tools work unchanged, but your requests never go to either
  company. They are served by the model you deployed.
* **Open-source software** such as vLLM, SGLang, Ollama, Unsloth, SkyPilot, and
  LiteLLM runs on Valkyrie's own infrastructure and does not send your data
  anywhere.

## Changes to this list

We update this page when we add or replace a subprocessor. See the
[changelog](/changelog) for dated changes.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.