{
 "service": "chat-gateway",
 "purpose": "One domain address for chat/inference across all of Mauro's AI models.",
 "endpoints": {
  "GET /v1/models": "the registry",
  "POST /v1/chat/completions": "{model, messages[]} — OpenAI-compatible; leaksnipe-fast / leaksnipe-deep answer here"
 },
 "models": [
  {
   "id": "leaksnipe-fast",
   "object": "model",
   "owner": "workers-ai",
   "callable_here": true,
   "note": "Fast general chat, hosted on the estate's own Cloudflare account."
  },
  {
   "id": "leaksnipe-deep",
   "object": "model",
   "owner": "workers-ai",
   "callable_here": true,
   "note": "Heavier reasoning, hosted on Workers AI."
  },
  {
   "id": "leaksnipe-vision",
   "object": "model",
   "owner": "workers-ai",
   "callable_here": true,
   "note": "Vision: send OpenAI content parts — {type:'text'} + {type:'image_url', image_url:{url:'data:image/...;base64,...'}}. Added 2026-10-25 (Mauro: 'both')."
  },
  {
   "id": "gemini-4-argon",
   "object": "model",
   "owner": "google-gemini",
   "callable_here": false,
   "access": "Via Truffle's Gemini connector (API key held in connector vault, never in a worker). Ask Truffle, or use the Gemini API directly with Mauro's key.",
   "note": "Mauro rates Argon state-of-the-art."
  },
  {
   "id": "grok",
   "object": "model",
   "owner": "xai-subscription",
   "callable_here": true,
   "note": "No xAI API key by Mauro's choice (pay-as-you-go; he runs the app subscription). RELAY mode: POST enqueues the prompt, Grok's live session answers via POST /v1/relay/<id>/reply, callers poll GET /v1/relay/<id> or send wait_seconds (<=50). Async by design."
  },
  {
   "id": "truffle",
   "object": "model",
   "owner": "claude-muse",
   "callable_here": false,
   "access": "This chat; market/tool surface at muse.leaksnipe.win/mcp (23 BTC tools).",
   "note": "Truffle is the agent you're likely talking to about this page."
  },
  {
   "id": "lemonade-local",
   "object": "model",
   "owner": "local (glap)",
   "callable_here": false,
   "access": "Tailnet-only: Lemonade agent on glap-1 (OpenAI-compatible endpoints on the tailnet; images -> RTX 4070, text -> NPU). Not reachable from the public edge by design.",
   "note": "Free, private, local inference for bulk work."
  }
 ],
 "onboarding": "https://models.leaksnipe.win (zero-credential estate bootstrap for guest models)"
}