We build the tooling that connects an assistant to your actual systems — your inventory, your servers, your mail — with permissions, a record of what it did, and control over who can ask for what. And we publish it in the open.
The assistant reads YOUR system instead of making things up. With permissions and a record of what it did.
Companies with their own systems that want to use AI without shipping their data anywhere and without signing a blank cheque.
You want to use AI but cannot send your data outside.
You tried an assistant and it knows nothing about your business.
Nobody can explain what the tool actually does with your data.
Every project is quoted to measure: first we understand what you need, then you get a closed price in writing. No catalogue rates.
The code is yours and auditable. Even if we stop working together, it lives on.
Exactly what it connects to and what it sends. No black boxes.
On your own servers if needed, with a demo mode to try without touching production.
The project picks the tool, not the other way round. If your shop runs fine where it is, we make it fast instead of migrating it for the sake of it.
Model Context Protocol: the standard an assistant like Claude or ChatGPT uses to talk to external tools. It is what lets it read YOUR system instead of making things up.
Only what you decide. The tool can run on your own infrastructure and every connection it makes is documented.
That is why it ships open with the documentation beside it: even if we stop working together, the code is yours and anyone can pick it up.
If it speaks MCP or has an API, almost always. We check on the call.
Yes: we publish several tools in the open. They are in Labs.
No strings. Whether or not you hire us, you leave with a concrete list of fixes for your site.
Book 20 min