SovrGPT Docs
API

API overview

OpenAI-compatible HTTP API on European infrastructure.

SovrGPT offers an OpenAI-compatible HTTP API. Existing OpenAI clients work without a single code change — just point baseURL and model at us.

Base URL: https://sovrgpt.com/api/v1

What is available?

EndpointStatusNotes
GET /v1/modelsstableLists every active model — chat and embedding, re-ranking, speech in and out.
GET /v1/mestableWhat is this key allowed to do? Every scope with a granted flag, no trial and error.
POST /v1/chat/completionsstableStreaming via SSE, or plain JSON.
POST /v1/embeddingsstableOpenAI-compatible · BGE-M3 & Nomic Embed Code, EU-sovereign.
POST /v1/rerankstableCohere-compatible · BGE reranker (cross-encoder).
POST /mcp/mcpstableModel Context Protocol — chat, search, image and video tools for Cursor, Claude Desktop, Zed, Windsurf.
POST /mcp/docs/mcpstablePublic documentation search via MCP, no API key required.
POST /v1/audio/speechstableAudio — text-to-speech: Supertonic (fast) or CosyVoice 3 (German, emotion, inline tags, voice cloning), self-hosted in the EU.
POST /v1/audio/transcriptionsstableAudio — speech-to-text (Voxtral, Paris/EU).
POST /v1/audio/uploadsstableAudio — upload ticket for recordings above 4.5 MiB: a signed address you upload to directly, limit 100 MiB.
POST /v1/images/generationsplanned Q3/2026Z-Image / FLUX.2 as REST.

Schema compatibility

We follow the OpenAI v1 schema as of 2026-04. Fields that exist in OpenAI but make no sense here (e.g. seed on some models, logit_bias) are accepted by the server and silently ignored — that keeps existing clients from crashing.

SovrGPT-specific fields (e.g. tier in the models response) are additional to the OpenAI schema and never break a standard client.

Rate limits

  • Standard plan: 100 requests/minute per API key, fair use on token volume.
  • Pro plan: 1,000 requests/minute, higher token quota.
  • Enterprise: agreed individually.

Headers on a 429:

Retry-After: 30
X-RateLimit-Limit: 100
X-RateLimit-Remaining: 0
X-RateLimit-Reset: 1715520000

Versioning

  • Current version: v1.
  • Breaking changes get a new version number (v2, v3, …) with at least six months of overlap.
  • Additive changes (new fields, new model IDs) are announced in the changelog: GitHub Releases.

Next

API overview