What it does

Game-Bot records your keyboard presses and mouse movements while you play any game, then trains a deep learning model to replay those same inputs. You run create_dataset.py to capture your gameplay, train.py to build the model, and ai.py to let the AI take the wheel.

The interesting bit

The “learning” here is pure behavioral cloning — no reward functions, no understanding of game state, just pattern-matching your reflexes. It’s imitation as automation: the bot doesn’t know why you clicked, only when.

Key highlights

• Built with Keras and TensorFlow; includes TensorBoard logging
• Cross-platform Python 3.6 project with minimal dependencies
• Three-stage pipeline: record → train → deploy
• Explicitly designed to work with “any game” via OS-level input capture
• Still actively developed (author’s note)

Caveats

• Only tested on Python 3.6.0; dependency versions unspecified beyond requirements.txt
• No mention of handling game timing, frame rates, or input latency — likely brittle in practice
• The README’s “deep learning” explanation is hand-wavy even by README standards

Verdict

Worth a look if you want a dead-simple intro to behavioral cloning or need to automate repetitive grinding in a static game environment. Skip it if you need adaptive AI — this bot won’t react to surprises you didn’t personally demonstrate first.