> ## 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.

# Inference API

> Call your deployed Valkyrie models over an OpenAI- or Anthropic-compatible API.

Valkyrie's public, client-facing API is the **inference API**: once you've
[deployed a model](https://docs.valkyrieapp.azumo.com/guides/deploy-model.md), you call it over a standard,
OpenAI-compatible (and Anthropic-compatible) endpoint from any client or SDK.

> [!NOTE]
>
> This is the only externally-supported programmatic surface. **Operating Valkyrie**,
> creating deployments, launching fine-tuning jobs, managing datasets, is done
> through the [dashboard](https://docs.valkyrieapp.azumo.com/dashboard/overview.md) or the [MCP server](https://docs.valkyrieapp.azumo.com/mcp/overview.md),
> not a public REST API. There is no public control-plane API to program against.

## Base URL

```
https://valkyrie-back.azumo.com
```

> [!NOTE]
>
> Use the API base URL for your environment. If your organization runs a separate
> environment, substitute its host. API keys are environment-specific.

## Endpoints

A ready deployment is addressed by its **slug**, `deployment-{id}` (for example
`deployment-227`), or by an [alias](https://docs.valkyrieapp.azumo.com/guides/aliases.md):

```
# OpenAI-compatible
POST /{deployment_slug}/v1/chat/completions
POST /{account_slug}/{alias}/v1/chat/completions

# Anthropic-compatible (Messages API)
POST /{deployment_slug}/v1/messages
POST /{deployment_slug}/v1/messages/count_tokens
POST /{account_slug}/{alias}/v1/messages
```

An alias points at a deployment or an inference pool. A pool alias serves only the
OpenAI-compatible routes, plus `GET /{account_slug}/{alias}/v1/status` and
`DELETE /{account_slug}/{alias}/v1/session`. Its Anthropic-compatible routes return
`400`. See [Point an alias at an inference pool](https://docs.valkyrieapp.azumo.com/guides/aliases.md#point-an-alias-at-an-inference-pool).

* **OpenAI-compatible** follows the Chat Completions contract, see
  [Chat Completions](https://docs.valkyrieapp.azumo.com/api-reference/chat-completions.md).
* **Anthropic-compatible** follows the Messages contract, so Claude clients work, see
  [Connect Claude Code](https://docs.valkyrieapp.azumo.com/connect/claude-code.md).

## Authentication

Send a **deployment API key** (`vk_dep_`). An alias that points at an inference pool
needs an account-wide key. Keys scoped to one endpoint get `403` there.

| Header | Notes |
| - | - |
| `X-API-Key: <key>` | Simplest for scripts and cURL |
| `Authorization: Bearer <key>` | What the OpenAI SDKs send, set the SDK `api_key` to your Valkyrie key |
| `x-api-key: <key>` | What the Anthropic SDKs send |

See [Authentication](https://docs.valkyrieapp.azumo.com/api-reference/authentication.md).

## Compatibility

Because the inference API is OpenAI- and Anthropic-compatible, you can use the
official SDKs, LangChain, and most third-party clients by pointing them at your
deployment's base URL. See [Connect a model](https://docs.valkyrieapp.azumo.com/connect/overview.md).


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