The research stack nobody built for sellers — and why Opyt is worth ten minutes of your week
Cross-border sellers don’t have a research problem. We have a retrieval problem. The average Amazon FBA brand owner I talk to is subscribed to a dozen newsletters, has three folders of saved TikTok Shop teardown videos, a Notion graveyard of “supplier to contact” notes, and a browser bookmarks bar that hasn’t been opened in months. The information exists. The problem is that when you actually need it — say, at 11pm the night before a Q4 restock decision — none of it is reachable inside the tool where you’re already working. Opyt, a small MIT-licensed tool launched on Product Hunt by David Maimon, is an attempt to fix exactly that. It’s not a seller tool. But the pattern it implements — turn trusted sources into a queryable knowledge base, expose it inside the AI client you already use — is one every operator running a multi-channel catalog should be stealing this quarter.
What Opyt actually does, and the problem it quietly solves
The pitch, stripped of the launch-day gloss: you tell Opyt which sources you already follow — X accounts, Substacks, papers, repos — and it builds a living knowledge base from them, refreshing as those sources publish. You then query that base from inside Claude, Cursor, Codex, or any MCP-compatible client. No new dashboard. No new app to open. The maker’s own framing is telling: he kept bookmarking posts on X, subscribing to Substacks, and saving papers he meant to read, and most of them stayed unread — so he built the tool he wanted.
The mechanics matter more than the marketing. When you ask a question, Opyt does top-k retrieval with hybrid search, then opens each post in the top-k and injects it into the model’s context. That’s the maker’s own description in the Product Hunt thread, responding to a question from Justin Rockmore about ranking. It’s a retrieval-augmented generation pattern — nothing exotic in 2025 — but the packaging is the interesting part. The knowledge base can stay local. It’s free and MIT licensed. You pay only for the AI tools you use and any OpenRouter usage. And there’s a sharing primitive: you can give a friend access to your knowledge base, search both with one question, see whose collection each result came from, and revoke access whenever you want.
For a cross-border seller, the closest existing analog isn’t a research tool at all. It’s the folder in your Notion called “Competitor Intel” that you update twice a year. Or the Slack channel where your VA dumps screenshots of Temu listings. Or the Helium 10 keyword list you exported in March and never revisited. Opyt isn’t competing with any of those on features. It’s competing on the retrieval surface — which is the thing that actually determines whether any of that intel ever changes a decision.
Why Amazon sellers should care more than Shopify ones
Here’s the asymmetry I keep coming back to. A Shopify DTC operator’s research loop is mostly creative and paid-media driven — you’re watching ad libraries, creative trends, TikTok Shop hooks. That work is visual, fast-decaying, and lives in Meta Ad Library and TikTok Creative Center. An external knowledge base helps, but the half-life is short.
An Amazon FBA brand owner’s research loop is different. It’s slow, compounding, and text-heavy: category reports, patent filings, supplier email threads, Seller Central policy updates, Jungle Scout and Helium 10 exports, Keepa price-history screenshots, Reddit threads from sellers who got suspended for the exact thing you’re about to try. That material should compound. It almost never does, because it’s scattered across six tools and two inboxes. A retrieval layer that sits inside Claude — where a lot of sellers are already drafting listings, parsing supplier contracts, and writing Klaviyo flows — is a much better fit for Amazon’s research cadence than it is for DTC’s.
How it differs from the tools you’re probably already paying for
Let me be blunt about the comparison set, because this is where most “AI research tool” launches fall apart.
The first category is note apps with AI bolted on — Notion AI, Mem, Reflect, Obsidian with plugins. These are storage-first. You have to put things in, and the AI helps you find them later. The failure mode is well documented: the input friction kills the habit. A commenter in the Opyt thread, Shivam Singh, described exactly this — he used Mind (per the thread) just to save things and find them later, and the shift Opyt represents is from storing posts to turning them into a knowledge base. That distinction is the whole product.
The second category is RAG-in-a-box for teams — Glean, Dust, Onyx. These are enterprise-priced, IT-deployed, and aimed at internal docs. A 12-person Amazon aggregator might justify Glean. A solo FBA seller with a VA in Manila absolutely will not.
The third category is the AI client itself — Claude Projects, ChatGPT custom GPTs, Cursor’s @docs feature. These are the real competitors. If you’re already paying for Claude Pro, you can build a rough version of Opyt with a Project, a few uploaded PDFs, and a weekly manual refresh. Opyt’s edge is that it keeps adding as your sources publish, and it works across multiple clients via MCP rather than locking you into one vendor’s memory feature. That’s a real architectural difference, not a feature-list difference.
The fourth category — and the one I’d actually benchmark this against for sellers — is the vertical research tools: Jungle Scout’s Cobalt, Helium 10’s Black Box, SellerSprite, DataDive. These give you structured category data. Opyt gives you unstructured source material. They’re complements, not substitutes — and honestly, the seller who wins is the one who pipes both into the same Claude conversation.
Where the math breaks
Opyt is free and MIT licensed, and you pay for the AI tools you use plus any OpenRouter usage. That’s the maker’s stated pricing model. Read it carefully, because “free” here means “you bring your own inference bill.” For a seller already on a Claude Pro plan, marginal cost is near zero. For a seller running heavy queries across a large knowledge base through OpenRouter, you’re paying per token — and RAG queries that open every top-k document in context are token-hungry by design. The maker confirmed this architecture directly in the thread. If your knowledge base is 5,000 posts deep and you’re asking broad questions daily, do the math before you commit. Not disclosed: any rate limits, storage caps, or what happens when a source you follow goes private or gets deleted.
What cross-border sellers can borrow from this, even if they never install it
The product is less interesting than the pattern. Here’s what I’d take from Opyt and apply to a seller operation this month, tool-agnostic.
Curate sources, not bookmarks. The Opyt model forces you to name the specific people and publications you trust. Most sellers do the opposite — they hoard links. Pick five to ten sources per channel (Amazon policy, TikTok Shop trends, Temu pricing, logistics, payments) and treat them as a standing feed. If a source hasn’t changed a decision in 90 days, cut it.
Put the knowledge base where the work happens. The single best design decision in Opyt is that it lives inside Claude, Cursor, and Codex rather than in a new tab. Sellers should apply the same rule to their own stacks: your supplier notes should be queryable from wherever you draft POs, not in a separate app you open on Fridays.
Use retrieval, not memory. The hybrid search + top-k + context injection pattern is the right one for seller intel because it keeps the model grounded in your sources rather than hallucinating category data. If you’re using ChatGPT or Claude for listing copy and category analysis, build the habit of attaching source documents rather than trusting the model’s priors.
Share the base, revoke the base. Opyt’s sharing primitive — grant access, search both bases with one query, see provenance per result, revoke anytime — is a clean model for how a brand owner should share intel with a VA, an agency, or a co-founder. Most sellers do this over Slack DMs and Google Drive links with no revocation story. That’s a compliance and IP problem waiting to happen, especially for sellers handling supplier contracts and unpublished product concepts.
The uncomfortable question: does this survive contact with a real seller workflow?
My honest read: Opyt is a beautifully scoped tool for knowledge workers who live in an AI client all day. Cross-border sellers are a subset of that group — the ones who’ve already moved their listing copy, supplier emails, and category analysis into Claude or ChatGPT. For that subset, Opyt is a genuine upgrade over manual Projects. For the larger group still running everything through Seller Central, spreadsheets, and WhatsApp, Opyt solves a problem they haven’t admitted they have yet.
That’s not a knock. It’s the same adoption curve every seller tool went through. Helium 10 in 2016 looked like a toy to sellers who swore by manual BSR tracking. The pattern is right. The timing depends on how deep into AI-native workflows your operation already is.
Where my judgment says it falls short
Three things I’d want before recommending this to a seller running real volume.
Provenance and auditability. The maker says you can see whose collection each result came from when sharing. But for a seller making sourcing or pricing decisions, I’d want per-claim citation — which post, which paragraph, which date. RAG systems that inject whole documents into context are notoriously bad at this. Not disclosed in the thread.
Source freshness and staleness. Amazon policy changes weekly. Temu pricing changes daily. A knowledge base that “keeps adding as they publish” needs a decay model — old posts shouldn’t rank equally with new ones for time-sensitive queries. The maker describes hybrid search but doesn’t describe recency weighting. That’s a gap.
No seller-native connectors. There’s no mention of Seller Central, Shopify Admin, TikTok Shop Seller Center, or any marketplace API. So Opyt can’t ingest your own operational data — only external sources. The most valuable knowledge base a seller could build is their own returns data, ad spend, and supplier history, cross-referenced against external intel. Opyt doesn’t do that today. That’s the version I’d actually pay for.
What I’d watch / test next
This week, before you install anything: audit your own research stack. List every source you claim to follow and mark the last time it changed a decision. I’d bet 80% fail the test. Then pick one channel — I’d start with Amazon policy and category intel, since that’s where the compounding is highest — and build a minimal version of Opyt’s pattern using Claude Projects with a weekly refresh ritual. If that habit sticks for two weeks, go install Opyt and let it automate the refresh. Watch three things over the next quarter: whether the maker adds recency weighting, whether any marketplace connector appears, and whether the sharing primitive gets provenance down to the paragraph level. If all three land, this becomes a default part of the seller AI stack. If only the first lands, it stays a power-user tool for the AI-native slice of operators — useful, but not yet load-bearing.
