Visual search should be a feature, not a project.
Trooply is a visual search platform for commerce. Every shop knows its shoppers are uploading photos of things they want to buy — TikTok clips, Instagram saves, street-style screenshots. Almost nobody has the infrastructure to turn that upload into a product page. We built Trooply so they don't have to.
What we make
A multi-tenant API that turns any catalog into a searchable vector index in minutes. Shoppers can search by photo, text, voice, crop, or a combination. Merchandisers run campaigns without paging engineering — pin a SKU during a sale, boost a brand, bury a discontinued line, expire a promo banner at midnight. Developers install a drop-in widget with one script tag, or build their own UI on top of the same API.
Under the hood it's CLIP ViT-L/14 for the retrieval layer, Qdrant for the vector store, Gemma 4 E4B for the natural-language and voice surfaces, FastAPI + Postgres for the platform. We didn't invent any of these pieces — we wired them together into something a small merchant can use the same afternoon they sign up.
Why it exists
The search experience shoppers have on retail sites in 2026 is, frankly, embarrassing. Typo-intolerant exact match. Category trees that don't match how anyone describes products. No path from "I saw this bag on Instagram" to "add to cart". Large retailers have built visual search teams — ASOS Style Match, Amazon StyleSnap, IKEA Place — because the retention lift is real. Everyone else has been priced out of the capability.
We're a small team. We think that's changing quickly. CLIP-class models are open. Vector databases have gotten good. Inference is cheap enough to serve at the long tail. The remaining friction is integration and operation. Our job is to remove that friction so every ecommerce platform — not just the hyperscalers — can ship visual search without it being a quarterly initiative.
A handful of principles that decide every trade-off.
Merchants ship the software, not us.
Every feature has a self-serve path — portal UI for merchandisers, API for developers. If a merchant has to wait for us to configure something, we've failed.
Catalog data belongs to the catalog.
Per-tenant Qdrant collections. No shared model training across customers. No data-resale business model. A merchant that leaves takes everything; the embeddings die when the subscription ends.
Explain every ranking decision.
Every result carries a _match_breakdown. Every moved result stamps its merchandising rules. Black-box search is dead; merchandisers should be able to click any card and understand why it ranked where it did.
Open schema, open pricing.
OpenAPI spec on every endpoint a developer might call. Prices on the pricing page, not behind a sales call. Clear usage limits per tier, documented rate-limit headers on every response.
Company
Trooply Inc. is an independent company. We are technology-led — founders are engineers who spent time adjacent to search and ML systems at other companies before starting Trooply. The team is small (intentionally), fully remote, and operates primarily across European and Asian timezones.
We do not take venture-dilution rounds at the size that forces hyper-growth narratives. We're building for a multi-year horizon and want to be around when merchants need a platform partner for the next decade of commerce UX shifts.
How to get involved
- Try it. The Free tier is generous enough to index a thousand products. Sign up.
- Read the blog. Short posts on the features we've shipped and why they look the way they do.
- Talk to us. Sales, support, security, press — /contact.
- Partner with us. Build an integration, resell in your region, or embed Trooply in another platform — email [email protected].