Put AI to work on the work you already do.
Herendil is a small consultancy for companies that want AI to do part of their everyday work. We build agents, run the models behind them and connect what your company knows so people and agents can find it. Then we hand it all to your team.
What we do
Agents. An agent is software that takes a task from start to finish. It reads the request, looks up the records, drafts the answer and asks a person before anything goes out. We start with one task you can measure, make it reliable, and only then add a second.
Inference. Which model runs, where it runs and what each answer costs. A small model close to your data often beats the largest one on the market for speed and price. We set up the serving, tests that check answers against your own examples, and a running tally of the cost per finished job.
The company brain. What your company knows is spread across documents, tickets, chat threads and a few people’s heads. We connect those sources, keep everyone’s access rights as they are, and make the result searchable by your staff and your agents. An agent is only as good as what it can look up.
How an engagement works
- A first call. Thirty minutes to see whether we can help. What it covers.
- Two weeks of study. We watch how the work gets done today, pick one task and write down what success looks like and how we’ll measure it. You keep the plan whether or not we go on.
- The build. We work in your systems, on your data, next to your team, and show progress every week. If the numbers stop improving, we stop and tell you why.
- Handover. You own the code, the prompts and the tests. We write it up so your team can run it without us.
What we won’t do
- Sell you a platform. We have nothing to license, so we gain nothing by locking you in.
- Start with a chatbot because it demos well. We start with a task someone does every day.
- Quote an accuracy figure we haven’t measured on your own examples.
- Keep your data after the work ends, or train anything on it.
- Take on a project we don’t expect to pay for itself.
Essays
- The company brain is a permissions problem
- Start with the boring task
- What an agent costs per finished job
- Small models, close to the data
Book a call
Bring one piece of work your team repeats every week, and tell us what goes in, what comes out and who checks it. We’ll ask where the data lives and what you’ve tried already. In thirty minutes you’ll learn whether AI is the right tool, where we would start and roughly what it would cost to run. Sometimes the right tool is a better spreadsheet, and we’ll say so.
Booking opens soon. The calendar will go here.