# Lynax CLIP Embedding Server

Base URL: `https://clip.lynax.ai`

Authentication: send either `Authorization: Bearer <CLIP_SERVICE_API_KEY>` or `X-API-Key: <CLIP_SERVICE_API_KEY>` on inference and model endpoints. Health and documentation endpoints are public.

Current model IDs:
- `openai_clip`: openai/clip-vit-large-patch14-336, dimension `768`
- `fashion_siglip`: hf-hub:Marqo/marqo-fashionSigLIP, dimension `768`

Stable integration endpoints:
- `GET /health`: public health check.
- `GET /llms.txt`: public AI discovery document.
- `GET /agent-docs`: public agent-oriented integration guide.
- `GET /agent/openapi.json`: public machine-readable OpenAPI schema.
- `GET /docs`: public Swagger UI for developers.
- `GET /models`: list enabled models, requires auth.
- `POST /embed`: multipart image upload using default model.
- `POST /embed-url`: JSON `{"url":"https://..."}` using default model.
- `POST /embed-urls-batch`: JSON `{"urls":["https://..."]}`, max batch `8`.
- `POST /models/{model_id}/embed`: multipart upload with selected model.
- `POST /models/{model_id}/embed-url`: URL embedding with selected model.
- `POST /models/{model_id}/embed-urls-batch`: batch URL embedding with selected model.

Default model: `openai_clip`.
Max image bytes: `8388608`.
OpenAPI for agents: `https://clip.lynax.ai/agent/openapi.json` (public).

Example:

```bash
curl -sS -X POST "https://clip.lynax.ai/embed-url" \
  -H "Authorization: Bearer $CLIP_SERVICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"url":"https://example.com/image.jpg"}'
```

Response includes `model`, `dimension`, `embedding`, and `elapsed_ms` for URL endpoints.
Do not mix vectors from different model IDs in the same Qdrant vector space unless that collection was designed/backfilled for that model.