Herendil helps companies put agents, inference and a shared company brain to work. We advise, and when building the software is the faster route, we build it.
Fig. 1What a company knows starts scattered across notes, tickets and threads, and lines up once it has somewhere to go.
Three areas. Most projects touch all of them, so we treat them as one piece of work.
Fig. 2One request, several routes, one checked result. Each ring is a tool call; the red route is the one that returned the answer.
An agent takes a task, finds a route through your tools and hands back a result someone can check. We design that route, the permissions it runs with and the points where a person signs off. Then we build it and run it with your team until it holds up.
Fig. 3Many small requests served side by side. Hatch density follows load; red cells came back from cache.
Choosing a model is the first step of many. We test candidates on your own data, decide where they run, and set up serving so that latency and cost stay inside limits you agree on before launch.
Fig. 4Every source passes through one knot and fans out to whoever asks.
Most of what a company knows sits in documents, tickets and people's heads. A company brain puts its knowledge, decisions and work in one place that people and agents can ask, under the permissions each source already has. We start with one team and one kind of question.
Small engagements, run in the open, with your team in the room.
One call about one problem. We decide together whether we are the right people for it.
We spend time in your tools and documents to see how the work gets done today.
We build the smallest version that does real work, in your stack, with your people.
You keep the code, the notes and the know-how to run it without us.
Essays and experiments, published as they are written.
Tell us what your company is trying to do and where knowledge gets stuck. If we can help, we will say how. If we can't, we will say that too.
30 minutes, by video