We build bots that don't trust each other.
aykol builds two autonomous systems: BrugBot, which makes decisions in the crypto market, and RedHorse, which reads football and basketball matches through numbers. In both, every container is a worker with one job; the rules live in code, and everything they do can be watched on screen.
- Every container has one job
- No rule goes live without evidence
- Every decision is on record
Two projects, side by side
One makes decisions in the market, the other reads matches. Different fields, the same way of building.
| Feature | ||
|---|---|---|
| Field | Memecoin and stock trading | Football (19 European leagues), NBA and EuroLeague |
| What it produces | Debated, audited buy proposals and position management | 6 explained coupons a day, a match board and a match advisor |
| Who decides | A twelve-agent desk proposes, an independent approver has the final word | Statistics and market pricing pick; you decide whether to play |
| Role of AI | Analysts, debate and a sceptical auditor | News reading, second-opinion review, the advisor and mascot commentators |
| How it measures itself | Score model, backtest, advisor and student; a 95% confidence interval requirement | Every pick is logged; ROI and closing line value (CLV) tracked |
| Monitoring screen | BrugBot Castle: dashboard, 2D map and 3D kingdom | Match board, sports newspaper and advisor line |
| Interface language | 7 languages: Türkçe, English, 中文, हिन्दी, Español, Français, العربية | The same 7 languages; foreign news is translated into the panel's language |
| Status | Used internally by the team | In active development, membership coming soon |
A rule left to an AI's good intentions is not a rule. That is why we write the limits into code and give each part only the authority to do its own job.
Recently added
Features added to both bots in November 2026.
Score model
Ranks candidates without spending a model request. In the group where it has proven itself, the strongest go straight to the approver and the weakest never reach the desk.
BrugBot3D kingdom
The castle was redrawn from scratch and now has a 3D view too: it sits in the middle of a kingdom with coastlines, winding rivers and a dragon's lair.
BrugBotFree model chain
If a model doesn't respond, the next one takes over automatically; which model answered and how long it took is recorded. The paid model is locked in code.
BrugBotThe panel in seven languages
The RedHorse panel now speaks seven languages too. Foreign news is translated in batches and cached, only when someone opens the panel in that language.
RedHorseFour commentator horses
Yıldırım, Alaz and Pamuk each read every match their own way and make one pick; in the daily bulletin, Altın picks the scorer of the day.
RedHorseGoals and cards coupon
Every day at 12:00: anytime scorer, player to be booked and total cards picks, with real bookmaker odds. The referee's average is factored into the expected cards.
RedHorseHow we work
The same six principles apply to both projects. Each one has a concrete example from a project.
Every container has one job
The scanner doesn't decide, the decider can't trade. If one part breaks, only its room closes.
BrugBot: the desk writes proposals, the approver executes, the vault can only sell.Job card
Every worker reads its card at start-up: its purpose, inputs, who it reports to and what it must not do. The contract part of the card is checked in code.
RedHorse: the advisor can't say "sure thing" or "lock", and can't make up numbers.Keys only where they're needed
Secrets live outside the project folder. Coding agents and analysis containers never see them.
BrugBot: the signing key exists only in the approver and the vault.No promotion without measurement
A new rule first runs in shadow mode; it doesn't go live until it is proven on time-separated data.
RedHorse: the analyst's "risky" note doesn't remove picks from a coupon until its accuracy is measured.Free and open data
Public sources and collectors that don't re-download unchanged data. If a source goes down, the bot carries on without it.
RedHorse: football-data, FotMob, ESPN and sports news feeds.A screen you can see inside
An office instead of logs: who's working, who's waiting, who crashed. The screen only reads; it can't give orders.
BrugBot: a crashed room's door shows "CLOSED" in the castle.From the blog
We write about why we build things the way we do.
- BrugBot
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. - BrugBot
Why the approver doesn't trust the desk
In BrugBot, twelve agents can agree on a buy and still no money moves. The final word belongs to a separate container that trusts the desk for nothing. Why we built it this way, and what it catches in practice. - BrugBot · RedHorse
How we run 12 agents on free model tiers
A twelve-agent debate sounds like dozens of model requests per analysis. In BrugBot an analysis averages about five. How early exits, collectors that spend no requests and a provider chain got us there.
Shared infrastructure
Both bots use the same AI connection: NVIDIA first, then OpenRouter's free models if it can't be reached. When AI is unavailable, both keep running without it.
- Python
- Docker Compose
- SQLite
- NVIDIA NIM
- OpenRouter
- TradingAgents
- GMGN
- football-data.co.uk
- FotMob
- ESPN
- Telegram
- Cloudflare