Why Your Cross-Border Team Needs a Shared Brain — and Not the Kind You’re Using
Every cross-border operator I know has the same pain: you’re running ads on Amazon, Shopify, and maybe TikTok Shop, you’ve got customer inquiries in three different inboxes, your warehouse is in a different time zone, and your AI tools are starting to outnumber your human teammates. The result is a fog of context that costs you directly — duplicated ad builds, conflicting inventory decisions, and onboarding that takes weeks instead of days. Most teams try to solve this with Notion, a shared Google Drive folder, or a prayer. That’s not enough when your agents are writing faster than humans can read. Compendium from Cerenovus (YC S26) is trying to be the missing layer: a persistent, automatically-updated shared context for both people and AI agents. It’s early, it’s rough, and it’s aimed at “AI-native tokenmaxxers” — but the core idea is exactly what any scaling e-commerce operation should be thinking about right now.
What Problem Does Compendium Actually Solve for Cross-Border Teams?
The pitch from co-founder Jonathan Waldorf is blunt: “If you’ve ever built a feature only to find that your teammate has built an identical one, or made an architectural decision only to have a teammate make a conflicting one, this product is for you.” In e-commerce, the equivalent is launching a duplicate PPC campaign because the person who already tested that keyword set left their analysis in a Slack thread that’s buried under three weeks of customer complaints. Or having your fulfillment team order 500 units of a SKU that your sourcing team just discontinued on the same day.
Compendium ingests information from email, Slack, and “basically everything else” — it’s a passive collection engine that builds a single source of truth for your team and your AI agents. That’s the first thing I’d steal even if you never touch the product: stop relying on humans to manually log context. Every tool you already use (Amazon Seller Central, Shopify admin, Klaviyo, Helium 10, your return portal) generates data that should feed a central knowledge base automatically. Compendium’s “always in the loop” feature — a live view of what every teammate and agent is working on plus summaries of what changed while you were away — is the kind of dream state that most operators only achieve after a painful post-mortem.
The multiplayer sessions are a closer second. The team describes it as “a Google Doc for AI” — two or more people collaborating in the same Claude session, with the shared context available in real time. I can see this being used to hash out a complicated Amazon policy change or jointly review ad creative across time zones. It’s a far cry from the solo AI sessions most sellers use today.
How Compendium Differs from Existing Options
The natural comparison is Notion. CTO Oliver Moreland is refreshingly honest in the comments: “I wouldn’t bill ourselves as a Notion killer (yet).” Notion has databases, views, public publishing, forms, and templates that Compendium hasn’t built. But Notion’s AI features are add-ons, not the core architecture. Compendium is built “AI-first” — structured notes exist, but the real value comes from passive information ingestion and agent-driven synthesis. For a GTM team, the difference is that you don’t go to a document to find information; the information finds you when you need it, via agents that answer questions from the ingested data.
Other “shared brain” tools like pumaDB, Unabyss, and Shram were mentioned in the comments for offering streamable HTTP MCP servers — meaning they aren’t locked to any one AI model. Compendium also uses a streamable HTTP MCP server, and Oliver confirms in the pricing thread that you can bring your own API key. That matters because many e-commerce operators are already locked into OpenAI, Anthropic, or Gemini, and you don’t want another tool that dictates your model choice.
The biggest differentiator for cross-border teams is provenance tracing — a feature not yet shipped but described in the comments: you can trace parts of the context back to the exact author (human or agent) and then ask an agent to review and scrub contributions from a departed team member. In e-commerce, turnover is high, and the “why” behind pricing decisions or supplier relationships often walks out the door. A system that can isolate and prune old context without breaking everything else would be worth the subscription alone.
Why Amazon Sellers Should Care More Than Shopify Ones
Let me be blunt: if you run a Shopify-only store, you probably have a cleaner data environment. You own your backend, your CRM, your email list. The context fragmentation is real, but it’s manageable with well-integrated apps. Amazon sellers, by contrast, operate inside a black box. Your PPC data lives in Amazon’s API, your inventory levels come from Seller Central, your customer feedback is trapped in Buyer-Seller Messages, and your account health metrics change faster than Congress. Feeding all that into a shared context that agents can query is a higher ROI proposition because the manual effort of reconciling Amazon’s fragmented data is enormous. Compendium’s passive ingestion could be the first time your Amazon account manager and your warehouse lead see the same version of “what happened last week” without a meeting.
That said, Amazon-specific integrations are not mentioned in the launch thread. If you’re an Amazon seller, you’ll need to check whether the 60+ integrations include Seller Central or Amazon Ads API. My guess is no — Compendium is targeting startups, not e-commerce. But the MCP server approach means a developer could build a connector. That’s a bet on future integrations, not a current capability.
What Cross-Border Sellers Can Borrow from Compendium’s Approach
You don’t have to sign up for this particular product (and I’ll get to the reasons you might not want to yet) to take away five operational principles that every scaling e-commerce team needs:
Centralize passively, not actively. Stop asking humans to write down what they did. That’s a doomed workflow. Instead, set up a system that monitors your key tools — Slack, email, your CRM, your ad platforms — and extracts decisions automatically. You can do this today with a combination of Zapier-to-Notion or a custom Slack bot, even without Compendium.
Give your AI agents persistent memory. Most sellers use AI for one-off tasks: write a listing, summarize a return. The next step is agents that remember previous tasks and decisions. The commenter Grey Seymour described his 15-20 page “soul canon” that he feeds to any new agentic surface. That’s the level of persistent context that Compendium aims to provide at team scale.
Run multiplayer sessions for high-stakes decisions. When you’re deciding whether to cut a 10% price on your best-selling ASIN, don’t let two people run separate Claude sessions that produce conflicting advice. Use a tool where you can both see the same context and prompts. Compendium’s multiplayer is a unique feature, but even a shared Google Doc with a running chain-of-thought helps.
Implement provenance tracking for departing teammates. Someone leaves your ad agency or your in-house marketing role. How do you know which ad copy decisions they were responsible for? Without provenance, you either lose that context or keep it and risk using stale information. Build a culture of tagging the author on every decision document, and invest in a tool that can trace contributions.
Use agent-driven periodic review. Gal Dayan raised the important question of what happens when context goes stale. Oliver’s answer — agents that periodically review context and update it — is the right pattern. Apply this to your SOPs: have an AI agent read your top 10 operational documents every month and flag anything that contradicts current reality.
Where the Math Breaks: Pricing and Maturity Concerns
The pricing thread is where this product reveals its early-stage identity. Originally, the team asked $100/month with no indication of what that included. After pushback from Grey Seymour, they clarified: $40/user/month (discounted to $20 with code LAUNCHDAY), which includes $40/user/month in Anthropic/OpenAI credits and 60+ integrations. For a team of five, that’s $100–200/month. That’s not expensive for a piece of infrastructure that could save you from one mis-shipped inventory order. But it’s not cheap either, especially for a product that the makers themselves admit is “two months old and we know it.”
The risk is that Compendium is still finding its defaults. Oliver says the periodic agent review is “opt-in, not plug-and-play yet.” The product is “canvas-first, opinionated-later.” If you’re not technical, you might spend more time configuring it than benefiting from it. The CTO’s honest answer — “we don’t have it locked down yet” — is refreshing but not comforting if you’re trying to run a holiday season on top of an untested context layer.
Also, no mention of Slack, email, or any e-commerce-specific integration in the list. The “60+ integrations” are presumably developer tools (GitHub, Linear, etc.). So unless you have a developer on your team who can build MCP server connectors for Amazon or Shopify, you’re getting a generic note-taking tool with fancy AI at the same cost as a Notion AI plan.
What I’d Watch / Test Next
I’m not suggesting you rip out Notion tomorrow and bet your Black Friday operations on Compendium. But here are five actions you can take this week:
Sign up for the 14-day free trial at cerenovus.app and use code LAUNCHDAY for 50% off. Don’t go all-in — just connect one or two data sources, like your Slack channel where ads decisions are discussed and your email inbox for supplier correspondence. See how the passive ingestion handles real messy data.
Test the multiplayer session with a colleague. Pick a current problem — e.g., “should we expand our SKU count by 30% next quarter?” — and run it through Compendium’s shared Claude session. Compare the output to what you’d get if you both ran separate sessions and compared notes manually. Does the shared context produce better decisions?
Check your current tool stack for MCP server compatibility. If you’re already using OpenAI or Anthropic, ask your developer if they can set up a custom MCP server that feeds your Amazon order data or Shopify product catalog into Compendium. If not, at least understand the architectural pattern so you can adopt it with whatever knowledge base you end up using.
Evaluate the provenance feature when it ships. This is the one feature I think could be a real moat for e-commerce teams. Ask the team for early access. If you can run a test where you trace an old decision back to a specific human author and then ask an agent to review its continued relevance, that’s a win.
Set up a manual equivalent today. Even if you don’t use Compendium, create a Slack channel called #shared-context or #team-decisions, and enforce a rule: every decision that affects pricing, inventory, or creative gets tagged with a date and author. Then hire an intern (or a cheap SDR) to copy-paste that into a running Notion doc. It’s primitive, but it’s better than nothing. Compendium automates what too many teams still do by hand.
This is a product to watch, not to adopt blindly. But the pattern — passive ingestion, persistent agent memory, multiplayer sessions, provenance tracking — is the future of cross-border operations. The first team that gets this right will have a compounding advantage over everyone still living in Slack threads and siloed spreadsheets.
