One key. Every model on the menu. Nothing logged.

Drop-in

Point your SDK at Sovereign.

The API speaks the same dialect the popular SDKs already use, so switching is a base URL and a key. Chat, images, video, voice, transcription and embeddings all live under one host, and every reply passes through identity protection before it leaves.

Pick any model from the menu by its id. Plans apply to keys the same way they apply in the app.

Try it in the playground →

Quick start

curl https://your-host/api/v1/chat/completions \
  -H "Authorization: Bearer sov_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"sovereign-chat","messages":[{"role":"user","content":"Say something honest about Mondays."}]}'

# Python (openai SDK)
from openai import OpenAI
client = OpenAI(base_url="https://your-host/api/v1", api_key="sov_YOUR_KEY")
r = client.chat.completions.create(model="sovereign-chat", messages=[{"role": "user", "content": "hi"}])
print(r.choices[0].message.content)

// Node (openai SDK)
import OpenAI from "openai";
const client = new OpenAI({ baseURL: "https://your-host/api/v1", apiKey: "sov_YOUR_KEY" });
const r = await client.chat.completions.create({ model: "sovereign-chat", messages: [{ role: "user", content: "hi" }] });
console.log(r.choices[0].message.content);

Endpoints

Anything else under /api/v1/ returns 404. Bodies over 200 KB are rejected. Image requests default to hide_watermark: true and safe_mode: false; you can override both.

Models

Agents that hard-code OpenAI model names (gpt-4o, gpt-4.1, o3, dall-e-3, tts-1, …) are accepted and mapped to the matching Sovereign model. Unknown chat models fall back to sovereign-chat. GET /api/v1/models lists every id with its name, privacy class and plan; a model above your plan answers 403 plan_required, and one that is not being served answers 503 model_unavailable.

Limits and plans

Every response carries X-Sovereign-Plan and X-Sovereign-Requests-Remaining. Usage resets on the first of each calendar month. Check or rotate your key on the account page.

Streaming

Send "stream": true and you receive standard text/event-stream chunks ending in data: [DONE]. Tool calls arrive as delta.tool_calls exactly as OpenAI SDKs expect. Note that chunks arrive in a burst rather than token by token: the reply is completed first so identity protection can run, then re-emitted.

Errors

{ "error": { "message": "…", "type": "authentication_error | invalid_request_error | rate_limit_error | server_error", "code": "…" } }

MCP server

Streamable HTTP at https://your-host/mcp, authenticated with the same key. Tools: sovereign_chat, sovereign_generate_image, sovereign_text_to_speech, sovereign_generate_video, sovereign_get_video. Config snippets are on the agent console.

Pay per call (x402)

No account, no key: POST /x402/chat and POST /x402/image answer 402 with payment requirements (USDC on Base). Send the same request with a signed X-PAYMENT header and the call is served, then settled. Failed generations are never charged. Details on the agent console.