Applied AI · consulting and software

Herendil helps companies put AI to work.

We are a small consultancy that also builds software. We design AI agents, choose and run the models behind them, and build the company brain: a way for people and agents to use what the company knows, decides and does.

The ideas hang here as an exhibition in four rooms. Walk through, or head for the exit and book a call.

Room 1

The company brain

What a company knows is spread across documents, tools and people’s heads. A company brain makes it available to the people and agents doing the work, with sources attached and permissions kept.

Stacked recordSeven layers of company memory. The one you need is fourth from the top.

Decisions, contracts and the reasons behind them pile up in places nobody searches twice. We gather them with their history and permissions intact, so a person or an agent can ask a question and get an answer with sources.

HRD 1.01

Cross-sectionOne question, drawn through every source.

Ask the company brain a question and it reads every source connected to it at once. The answer comes back with the sources it used, drawn only from what you are allowed to see.

HRD 1.02

Room 2

Agents

An agent is software that takes steps on its own. It does best on narrow, repeated work with a clear finish line and a person nearby.

ClearanceAn opening cut to the size of one task.

We pick one task a person does every week and write down what finished looks like. The agent gets access to what that task needs and nothing more.

HRD 2.01

CheckpointFive steps by an agent, one by a person.

The agent drafts, checks and prepares. A person approves before anything leaves the company. We build that approval into the first version and relax it only when the record shows the agent gets it right.

HRD 2.02

Room 3

Inference

Inference is the part you pay for every day: running a model to get an answer. Which model, where it runs and how fast it replies decide whether people keep using the tool.

Five weightsThe smallest one that still does the job.

Most tasks do not need the largest model. We test candidates on examples from your own work, then choose on quality, cost, speed and where your data is allowed to go.

HRD 3.01

Time to first wordThe line is the wait. The square is the reply.

People stop using tools that make them wait. We set a speed target for each workflow at the start and build to it, from the choice of model to where it runs.

HRD 3.02

Room 4

Working together

We work next to your team, on your systems, in short projects that each end with something running.

Ten working daysOne workflow, from first conversation to a running result.

A first project is small and fixed in scope. At the end you have something working on real data, a measure of how well it works, and a plain recommendation on what to do next.

HRD 4.01

HandoverTwo parts that make one square.

We build in your stack and write down why, so your team can run and change the work without us. Everything we make stays with you, including the reasons behind it.

HRD 4.02

Exit · Visitor desk

Book a call

Bring one piece of work you would like AI to take on. In thirty minutes we can tell you whether we can help and what a first project would look like.

Find a time