There are two kinds of provider in this field. One explains what you could do with AI. The other operates something with it and therefore knows where it breaks.
The difference does not show in the sales conversation, it shows eight weeks later. When a system runs for four months straight, problems appear that are in no manual: storage grows and nobody clears it. A model suddenly answers differently than before. An access token expires, and the failure does not arrive as a failure but as a quietly wrong result.
This series describes what runs at our place. Not as a product catalog but as a workshop report.
Why We Build Our Own At All#
The honest answer is not ideology but a series of moments where something existing was not enough.
An assistant that remembers nothing is useless in daily work. So we needed a memory. A memory that only grows becomes a haystack within six months. So we needed clearing, weighing and forgetting. Agents running the same job every night do not get better by themselves. So we needed a mechanism that measures their instructions and swaps them.
Every piece in this series came out of a moment like that, and some of them turned into products of their own.
What There Is#
The memory. The core of everything. A system that keeps conversations, decisions, people, projects and connections across months and, on request, returns exactly the three things that matter right now. It runs for ourselves, for clients, and as a service others can subscribe to. It is the oldest and most-used piece in the house.
Darwin. Agents that improve their own instructions. Every run is scored, a challenger text is derived from the results, both run against each other, the better one stays. With safety gates, so a bad challenger never goes live. It is open source and by now our most-downloaded package.
MeetMyAgent. A platform where anyone becomes visible for free, to humans and to AI. The idea sits in the name: every profile introduces its agent, and that agent acts on behalf of its owner. Where money is involved a human still decides.
The Academy. Our learning platform for memory-first AI, MCP servers and agent patterns. What makes it interesting is not the content but the operation: it is run by a dedicated group of agents that propose, review and write content and watch its visibility.
Three self-serve tools. A contact system that lives inside a chat window rather than an interface. A tool that lets several agents split one job between them. And one that measures whether and how a website appears in AI answers.
The agent fleet. The largest group and the most invisible one. Agents that check, gather and compare at night and leave a report in the morning. For us and for client sites.
The working system itself. The editor, the models beside it, and the layer of rules, recipes and guard rails that stops a well-meant automation from breaking something. That is the part talked about least and carrying the most.
The One Rule That Holds Everywhere#
If this series has a single line running through it, it is this: the assistant prepares, the human decides.
No system here sends a mail to a client on its own. None closes a contract. None deletes anything important without asking. That is not caution on principle, and not modesty about what would be technically possible. It is the realisation that the value sits in the preparing and the risk in the triggering. Give away the preparing and keep the trigger and you gain almost everything while risking almost nothing.
The second line is less comfortable: we are our own hardest case. Everything here runs at our place before it runs at a client's. That also means we see the uncomfortable findings at our own place first. Which is exactly why we know what to look for elsewhere.
What Is Not In This Series#
No client data. No credentials, no server internals, no excerpts from running systems. Where something involves a client, the company name and what we built is all there is.
And no numbers without a source. Where a number appears here, what it refers to appears with it. That makes some sentences longer than they need to be. The alternative is a language model passing on a claim of ours two months from now that nobody can back.
Where It Starts#
The first piece in the series is the one about memory, because everything else stands on it. After that come the agents, the platform, the learning site, the tools, and finally the working system in which all of it is built.
For anyone wondering whether we know what we are talking about: this series is the most honest answer we can give.
