EstuaryThe agents live where your data already does.
Everyone is being sold AI. Almost nobody is being shown where to point it. We are building a platform for small, pluggable agents that work on structured product logic instead of guessing at it: running inside your own environment, reaching your systems over an open protocol, and never asking you to hand your product data to anyone.
Read it from the outside in. Your environment holds the platform, the platform holds the agents, and the agents reach your systems through a connector layer: the Model Context Protocol where a vendor publishes one, a plain API adapter where they do not. Your systems stay exactly where they are, and nothing crosses the outer line.
The first four agents
These four are in build. The order they arrive in is set by what the waitlist asks for, so if one of them is the reason you are still reading, say which and it moves up.
Configuration Validator
In buildCatches the impossible combination while it is still a quote.
Checks a quote or order configuration against your product rules before it moves forward, so an invalid combination is stopped at the source instead of discovered on the shop floor.
BOM Reconciler
In buildFinds the line where PLM and ERP stopped agreeing.
Compares bills of material across systems and flags where they have drifted apart, before that drift reaches production and turns into a rework order.
Part Classifier
In buildTurns a catalogue back into the families it always was.
Groups parts across your catalogue by real similarity, surfacing duplicates and rationalization candidates that manual review misses because nobody can hold ten thousand part numbers in their head.
Change Impact Assistant
In buildShows what an approval is about to touch.
Traces what an engineering change actually reaches across product structure, quotes and orders in flight, before it gets signed off rather than after.
Something else on your list?
The agents that get built are the ones people keep asking for. If the job you want done is not on this page, that is worth a conversation on its own.
Tell us what it isHow it connects
Each agent is delivered as a tool your own AI client can call, using the Model Context Protocol, the emerging open standard for exactly this. You install it, point it at your systems and hold your own keys. There is no Estuary Logic platform sitting between you and your product data, and no rebuild of your ERP to accommodate one.
Going the other way, into your systems, the protocol depends on the vendor. Some PLM and CPQ products already publish an MCP server and an agent can speak to them directly. Plenty of ERP and CRM products do not, and for those we build an adapter against the API they do have. Either way the agent sees the same structured model, which is the part that matters and the part we own.
That has a consequence worth stating plainly: an agent is only as good as the structure underneath it. Which is why the fastest way to find out whether any of this will work on your product data is the free assessment, and why it takes an hour rather than a quarter.
Start with the free assessmentTwo things every agent has to do
Explainable by construction
Every answer shows which systems and fields it used, and links back to the source. Roles and rights are respected and everything is logged. An engineer should be able to check the work, not just trust it. An agent that cannot show its reasoning is not finished.
Runs in Europe, on your terms
EU hosting and GDPR by default. Agents run inside your own environment, with your keys and your choice of model, so your product data does not have to leave it. Nothing is trained on your data.
Get in before the queue forms.
The platform is where these agents live, next to the practical material that makes them work: how to structure what you already have, what to automate first, and what to leave well alone. Access opens in groups rather than all at once, and this list is how we choose. Put your name on it and tell us what you would want it to do first. That is what decides what we build next.