README.md

---
title: Afterimage API
emoji: 🛰️
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
license: mit
---

# Afterimage — live Qdrant backend

The real FastAPI + FastEmbed CLIP + embedded **Qdrant** backend behind
[Afterimage](https://afterimage-qdrant.vercel.app) — searchable visual memory for
physical spaces.

Every request runs live against Qdrant:

- `POST /api/anomaly/scan?scenario=vault` — `query_points` filtered nearest-neighbour;
  flags the region when its live crop falls below the data-derived floor (`mean − 3σ`).
- `GET /api/outliers?scenario=vault` — `RecommendQuery(best_score)` outlier vs baselines.
- `GET /api/matrix?scenario=vault` — `search_matrix_pairs` distance matrix.
- `GET /api/text_search?q=a+van` — open-vocabulary CLIP text→image search.
- `GET /api/health` — collection + embedding mode.

512-dim cosine vectors, one `object_memory` collection, no training, no labels.
Interactive docs at `/docs`.