API Reference

OpenAI-compatible API endpoints, authentication, chat completions, responses, streaming, and tool calls.

Sovereign exposes an OpenAI-compatible API under:

https://api.sovrun.one/v1

Authentication

Send your API key as a bearer token:

Authorization: Bearer sk-sov-live-...

Models

List available models:

curl https://api.sovrun.one/v1/models \
  -H "Authorization: Bearer sk-sov-live-..."

The stable public model name is:

sovereign-alpha

Chat Completions

Endpoint:

POST /v1/chat/completions

Example:

curl https://api.sovrun.one/v1/chat/completions \
  -H "Authorization: Bearer sk-sov-live-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sovereign-alpha",
    "messages": [
      {
        "role": "user",
        "content": "Create a concise checklist for reviewing firewall ACL changes."
      }
    ]
  }'

Streaming

Set stream to true:

{
  "model": "sovereign-alpha",
  "stream": true,
  "messages": [
    {
      "role": "user",
      "content": "Draft an incident report outline."
    }
  ]
}

Sovereign returns server-sent events compatible with OpenAI-style clients.

Responses API

Endpoint:

POST /v1/responses

The Responses API maps compatible requests into the chat completion pipeline.

curl https://api.sovrun.one/v1/responses \
  -H "Authorization: Bearer sk-sov-live-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sovereign-alpha",
    "input": "Explain the risk of exposed admin panels in a production network."
  }'

Tool Calls

Sovereign passes OpenAI-compatible tool definitions to the upstream reasoning layer.

Supported fields include:

tools
tool_choice
parallel_tool_calls
legacy functions
legacy function_call

Tool execution remains the client's responsibility. Sovereign forwards tool calls; it does not execute arbitrary client tools.

Unsupported Inputs

Image inputs are rejected with:

{
  "error": {
    "message": "unsupported_image_input",
    "type": "invalid_request_error",
    "code": "unsupported_image_input"
  }
}

Sovereign currently does not expose a multimodal vision model.

Errors

Errors follow the OpenAI-style shape:

{
  "error": {
    "message": "insufficient_credits",
    "type": "invalid_request_error",
    "param": null,
    "code": "insufficient_credits"
  }
}

API Reference

On this page