— your entire Amazon catalog on a visual canvas. AI photo pipeline, compliance checks, and direct SP-API push through your own credentials.
Four phases. The fourth is keeping the first three honest, forever.
Connect Seller Central with your own SP-API app. Press the button. The entire catalog lands in our database.
Bulk-select 50 SKUs, pick a template, hit Run. Compliant main + 6 lifestyle alts per SKU.
Compliance-check then write to your real listings via SP-API. No CSV upload. No browser robot.
If Amazon silently swaps your photo for a sizing grid, we ping you. One click to re-push the original.
Register your own private SP-API app inside Seller Central (free, five minutes). Paste your CLIENT_ID and CLIENT_SECRET into Oxus. We never see your Amazon password.
Pick "Amazon Compliant Pack" or "Amazon Alt Pack". The canvas runs your photos through background-remove, white-background standardize, 85% fill, and a final compliance gate. Inspect before you push.
Direct write to your listings via SP-API. We log every push, remember the workflow that made each image, and watch the live listing afterwards. If Amazon's AI Bot swaps your photo, you know within hours.
Six primitives. The whole product is in this grid.
Every SKU, every status, every AI run that ever touched it. Filter, bulk-select, run a workflow on any slice. Excel disappears.
Flux, z-image-turbo, Qwen-Image-Edit on a private RunPod endpoint. Open weights, no vendor C2PA tags, no model training on your data.
White background + 85% product fill + no text overlay + C2PA strip. The four things that get listings rejected — checked before push.
Your own SP-API app, your own refresh token, encrypted per workspace. We never see your Amazon password. Same direct channel Amazon's own integrations use.
Amazon's AI Bot quietly swaps your photos for AI-generated sizing grids. We poll the live listing every N hours and ping you the moment it changes. One click to re-push.
Every listing row remembers which workflow run produced its current image. Replay it, diff it, audit it. Nothing happens in the dark.
Real photos in. Compliant listings out. Pushed straight to Amazon — through credentials you own.
— the seller with a real product, a real warehouse, real photographs. Two days per new SKU bouncing between Photoshop, ChatGPT, Seller Central, re-shoot, re-upload.
You run your own warehouse. You take the photos yourself. Oxus turns each shoot into a compliant main + six lifestyle alts, ready for Seller Central.
Excel collapses at this size. The Listings DB is built for it — filter, bulk-select, run a workflow on any slice of the catalog.
Multi-workspace, white-label, audit-logged. On the v1 roadmap — talk to us if you need it sooner.
One pack = one SKU run through one template. Main + six lifestyle alts, or a branded pack — both count as one. Roughly the price of one freelance image gig.
Try the canvas on a handful of SKUs. For sellers checking before they commit.
For sellers running daily refreshes across 5 – 500 SKUs.
For agencies running marketplaces for clients, and high-volume catalog operators.
Run on a private RunPod endpoint. The same architecture that produced our pilot's 12k SKU images — proven, fast, no vendor C2PA tags Amazon now detects.
No C2PA. No vendor lock-in. Open weights run on our RunPod endpoint. Your data never trains anyone's model. Your generations carry no metadata that Amazon's 2026 detection pipeline can flag.
One SKU, one template run. Amazon Compliant Main + 6 lifestyle alts is one pack. Re-running the same SKU through a different template (say, a branded alt pack on top of the compliant one) is another pack. Replaying a previous run from provenance is free.
Most image generators are template-based — no visual workflow, no awareness of your catalog. Hosted image-gen services typically stamp C2PA metadata on every output, and Amazon's main-image AI detection picks it up. We run open-source models on our own RunPod endpoint, with no C2PA. And we push directly to your Seller Central through your own SP-API credentials. Canvas + catalog + open-weights AI + direct SP-API — four things, one product.
You register your own private SP-API app inside Amazon Seller Central — free, takes five minutes. You paste the CLIENT_ID and CLIENT_SECRET into Oxus. You authorize. The refresh token gets Fernet-encrypted and stored per workspace. Your Amazon password never leaves Seller Central.
Amazon's 2026 policy strongly favours sellers who own their own SP-API app. It sidesteps our 10-authorization cap, gives you full control, and means Oxus never sees your Amazon login. It is the more conservative, more compliant, more seller-friendly default — and it scales without an enterprise contract.
Our compliance node enforces the four big rules — white-only background, 85%+ product fill, no text or watermark, no C2PA metadata. Many hosted image-gen services skip the last one (C2PA gets stamped on every generation) — that's the most common reason for AI-generated images getting flagged in 2026. We run the check before every push.
Europe. Postgres on Hetzner in Falkenstein, Germany. Image storage on Cloudflare R2 (EU region). Inference on Nebius (EU) and a private RunPod endpoint. We don't train on your data. SOC 2 Type II in progress.
Amazon US and EU at launch — .com / .co.uk / .de / .fr / .it / .es. Same SP-API, the marketplaces differ only by marketplace_id. Shopify Admin is on the v1 roadmap. eBay v1.5. Walmart on demand.
Oxus started as a creative engine for DTC marketers — research trends, build ad creative on a canvas, publish, measure. It worked. But the sellers I knew kept asking the same off-script thing: "can you also build me an Amazon listing pipeline? a price-scraper? a compliance check?"
The pattern was unmistakable. Every Amazon seller has a Fiverr backlog of one-off image gigs they keep meaning to commission. That backlog is now the product.
Same canvas, same nodes, same provenance — focused: Amazon listings, end-to-end. I'm dogfooding it on my own car-parts shop, twelve thousand SKUs across Italy and the EU.
— your full Amazon catalog, on autopilot, in an afternoon.