How the Poker Bot Works

The bot is a loop that runs several times per second: look at the table, work out the game state, decide, act, log. Each step is described below.

1. Capture

The bot takes a screenshot of the table window – an emulator window for club apps or the room’s own client on desktop. It does not read memory or intercept network traffic; everything it knows comes from pixels.

2. Recognise

A calibration profile for each room tells the bot where cards, stacks, pot and buttons appear. Cards are matched against templates; numbers are read with OCR and sanity-checked against the previous state. If recognition confidence drops – an animation, a pop-up, a changed layout after an app update – the bot pauses rather than guess.

3. Decide

Preflop, the bot looks up the action in a range table keyed by position, effective stack and prior action. Postflop, common spots are also lookups; the rest go through a fast evaluator that estimates equity against an assumed range and picks the option with the best expected value under a simplified strategy. Decisions are deterministic given the same input plus a seeded random element for mixed strategies.

4. Act

The chosen action is executed through simulated input in the emulator or client, with a delay drawn from a configurable distribution so that the timing is not uniform. Bet sizes are rounded to amounts a human would use.

5. Log

State, decision, confidence and timing are appended to a local log. This is what you review after a session and what you send us when something looks wrong.

What this design means

Screen reading makes the bot independent of a room’s internals, but sensitive to layout changes: an app update can break a profile until we ship a new one. It also means the bot can be detected by the same means any client-side automation can. That trade-off is discussed on the detection page.