Skip to content
aykol.fyi
← All posts

BrugBot3 min read

Why we built a castle

BrugBot's dashboard is a pixel-art castle: every container is a room, and every agent is doing what it is actually doing right now. It looks like decoration, but it's a monitoring tool. Why we look at a castle instead of reading logs.

When you open BrugBot, you don't get a chart or a table. You get a walled pixel castle in front of the mountains. In the Seers' Room the analysts are writing, in the Divan Hall the bull and the bear are arguing, at the Seal Gate the approver is waiting, and in the Treasury open positions sit as piles of gold. At first glance it looks like a game. In practice it's the fastest way to read the state of the system.

Logs don't lie, but they talk a lot

In a system with this many containers, dozens of log lines scroll by every minute. To notice something going wrong, you have to dig the answers out of those lines: when did the desk last run, why did the approver go quiet, is the news crier sleeping or did it crash? Doing that every day is tiring, and it's easy to miss the most important signal of all: nothing happening.

In the castle, the same information is visible at a glance:

  • Every container is a room. The figure in the room is doing that container's current job.
  • A crashed service shows CLOSED on its door. You don't need to read logs to notice a part that has quietly stopped.
  • Open positions, the number of analyses and profit/loss stay visible at the top of the screen.
  • The castle ledger records proposals, approvals and rejections in order, with their reasons.

The screen only reads

The castle's most important feature is what it can't do. The screen only reads: it sees no keys, can't give orders and can't start a trade. Even if the interface had a bug, or someone got access to the screen, the system's behaviour wouldn't change. The emergency stop isn't on the screen either; it's a separate command or a Telegram message, and the screen only shows the result. The castle gates close and a red warning appears at the top.

Dashboard and map

There are two views, each answering a different question:

  • The dashboard answers "what happened today?". Search a token, open its analysis, and see the decisions of the desk, the approver and the vault in one file, with their seals.
  • The map answers "what is happening right now?". Who is working, who is waiting, who has stopped.

The interface comes in seven languages. Only the interface text is translated; reports written by agents, logs and news headlines stay as they are, so translation costs no extra model requests.

Brug and Seryali

Two more characters live in the castle. Seryali walks the rooms cheering the team on. Brug has no job at all; he wanders around and makes odd comments about what's going on. AI writes their lines from recent decisions and news, using very few requests.

The characters have a purpose too: the hardest part of monitoring a system is opening it regularly. Wondering what Brug said today is a good excuse to open the castle.

See for yourself

A copy of the castle running on sample data is on this site: open the BrugBot demo. Token names are made up, and there are no real trades or recommendations. But the rooms, the agents and the castle ledger are the same as in BrugBot's own interface.