One harness auto‑orchestrates a team of specialist AI agents across Amazon, Walmart, TikTok Shop, Bol, Allegro and Kaufland. It tracks every product, keyword and competitor all the time, and tells you what changed and why.
Competitor sales are increasing. Nobody can say which keywords, which images or which ads they changed.
You hear about a rank drop from your own weekly report, or worse, from the client.
Six marketplaces, dozens of accounts, hundreds of products. Who actually looks at all of it every day?
Connect your own Anthropic or OpenAI key and pick the model per agent, or use ours at cost. Your prompts, your usage data, your bill. No markup on compute.
Every action passes a confidence gate. Above it, the agent acts inside the limits you set. Below it, it asks. Not a mode switch; a decision at a time.
For every action type you decide who signs off: the AI, your team, or the client. Per account. An agency and its client can share one workspace with different rights.
What the agent saw, what it proposed, what you changed, what ran. One‑click rollback, one‑click export. A log you can hand to a client or a CFO.
A price move on Amazon DE changes your Kaufland buy box. Stock committed to TikTok Shop is stock Bol cannot sell. One catalogue, six rulebooks, read as one.
Plug our MCP into Claude or ChatGPT and ask about your accounts in plain language. The difference from Amazon's own MCP: we store. Ads data stays past the two months Amazon shows you; sales, fees, inventory and rank history stay as long as you want; and you connect every account you run, not one. Free, and you can sign up today.
The same data as a screen for the people who do not type prompts: sales, every cost line and margin per product and per market. Pick the columns, the markets and the period and export a PDF for your manager or your client. Free with the MCP.
The watch goes live. Keywords and competitors tracked all the time, every change explained, every finding lands in your action log for review. Free for the first users.
The agents start doing, not only watching: listings, ads, logistics and compliance tasks, each one on the autonomy level you set.
The same watch, the same rules, on Walmart, TikTok Shop, Bol, Allegro and Kaufland. One catalogue read across all of them: a price move on one marketplace shows up as a buy‑box risk on another.
Every action type sits in one of three columns: acts on its own, proposes and waits for your approval, never touches. Drag a card and the rule changes for that account. Start cautious, loosen as trust grows.
The first release watches and explains. Execution follows, always on your approval: approve, apply in one click, roll back. The agent never changes a listing on its own.
Ten years in Amazon and marketplace management, running an agency team that manages brands across European marketplaces today. Every account we managed by hand, every incident we resolved, is what the harness learns from.
Amazon EU and US first, then Walmart, TikTok Shop, Bol, Allegro and Kaufland. Tell us yours in the form; the list decides the order.
Not in the first release. It reads, watches and explains. When execution comes, nothing runs outside the permission level you set, and every action can be rolled back.
Through the official marketplace APIs (Seller Central login takes about two minutes) and through our MCP for your own AI assistant. Read access is enough for the watch.
Yes. Connect your own Anthropic or OpenAI key and choose the model per agent, or use our pooled models at cost. Either way you see what each agent spends and can cap it.
Only your workspace. Accounts are separated per client, each client can be given its own view, and nobody outside your team sees them. Your own key means your prompts never leave your account.
Yes. One workspace, different rights: you decide which actions your team approves, which the client approves, and which run on their own.
Each product gets a pool of keywords and a pool of competitors. You approve both; the agent proposes additions when it sees something new.
First users in December 2026. We write personally, in the order of the list.
The MCP and the analytics dashboard are free. The watch is free for the first users. Pricing for the full agents comes with the May release.
Built for teams running many accounts. If you run one brand with volume, it works the same way.
