Skip to main content
This section is about using a Valkyrie-hosted model for inference, pointing a client at a model you’ve deployed so it answers your prompts.
Want to operate Valkyrie from an AI agent instead, deploy models, stop them, read logs, manage jobs? That’s a different integration: the MCP server. Connecting a model (this section) and connecting the MCP server (that section) are separate things and don’t depend on each other.

Two compatible protocols

Valkyrie exposes your deployed model over two protocols, so most clients work out of the box:

OpenAI-compatible

The Chat Completions API. Works with the OpenAI SDKs and any OpenAI-compatible client, Cursor, LibreChat, editors, custom apps, and more.

Anthropic-compatible

Valkyrie can also speak the Anthropic Messages protocol, so Claude Code can point at a Valkyrie-hosted model.

What you’ll need

  • A ready deployment (or an alias pointing at one).
  • A deployment API key (vk_dep_) with access to that model, see API Keys.
  • Your API base URL (e.g. https://valkyrie-back.azumo.com).

Pick your client

Claude Code

Point your coding agent’s model at a Valkyrie deployment.

Cursor

Add a Valkyrie model as a custom OpenAI-compatible model.

OpenAI-compatible clients

The generic recipe for any OpenAI-compatible SDK or app.
A self-hosted open model is not a frontier model. Behavior, tool-use quality, and context length depend entirely on the model you deployed, size the deployment for the workload you expect.