Precision Control
An aimbot that runs on your GPU.
YOLO AI detection on a local crop. TensorRT .engine or ONNX. Windows only.

Crop. Detect. Lock. Move.
The whole product is this loop on your PC. No game process. No cloud inference.
Capture
Center crop via BetterCam or MSS. Nothing injects into the game.
Detect
YOLO AI detection draws boxes on the crop. Load a .engine or a .onnx. Raw .pt is not used live.
FOV
One target inside a circle or square. HSV and friend/foe if the model has it.
Mouse
Win32, GHUB, Arduino, spoofed Arduino, Makcu, or RP2040.
Features
What ships in the app.
Same feature set on every plan. Duration is the only difference. Activate the key in the desktop app.
- 01
Game packs and profiles
YOLO file sits on the game. FOV, aim keys, smoothing, and macros sit on the character. Switch in-game with Alt+1 through Alt+9.
- 02
TensorRT or ONNX
Per-game model path. Fastest is a TensorRT .engine on this GPU. Portable path is ONNX. Live inference never loads a raw .pt.
- 03
Aim, FOV, HSV
Sensitivity, smoothing, deadzone, adaptive tracking. Circle or square FOV with kill and capture zones. Optional HSV color gate.
- 04
Three aim methods
normal while the aim key is held. hold_release with optional wiggle. auto_trigger when the crosshair is inside the radius.
- 05
Mouse backends
Win32api, Logitech GHUB, Arduino, spoofed Arduino, Makcu, or RP2040. Serial protocol and COM port when you use a board.
- 06
Macros
Shared library of taps, delays, and key down/up. Enable which macros run on each character. Triggers are keybinds only.
- 07
Live preview
Detection boxes, FOV outline, zone rings, class labels, HSV mask, and FPS while the loop is running.
- 08
Import and export
Share a character, a whole game, or the macro library. License keys, model paths, and COM ports stay off the file.

Game packs. Character profiles.
FOV, aim keys, smoothing, and macros sit on the character. The YOLO file sits on the game pack. Switch in-game with Alt+1 through Alt+9.
YOLO AI detection. Two file types.
ClawAim does not inject. It crops the screen and runs a YOLO object detector so it can see players as boxes. That is the AI detection step. You pick one weights file per game pack: TensorRT .engine or ONNX .onnx. Live inference never loads a raw .pt.
.engine
TensorRT engine
Fastest path. Built for this NVIDIA GPU and this TensorRT version. Needs CUDA plus TensorRT (ClawRT). Cannot be copied blindly to another PC. If an engine and an ONNX exist, ClawAim prefers the engine.
.onnx
ONNX Runtime
Portable YOLO export. Import in Setup. Runs without TensorRT. Slower than an engine on the same card. Subscribers start here. The installer does not bundle models. You can build an engine from ONNX later if TensorRT is installed.

Which games use which file
Format is not locked to a title. Each game pack has its own model path. Either file works on any pack. Example packs in this repo:
- Marvel Rivals
- Example pack points at a TensorRT .engine (YOLOv8 weights exported for this GPU). Friend/foe class models also live as engines or ONNX plus a .classes.json sidecar.
- COD / Warzone
- Example pack points at an .onnx (for example WZ_50k at 320). Use this when you want a file you can share or run before TensorRT is installed.
- Any other pack
- Default starts empty. Import ONNX for that game, or drop an engine. Switch games and import again if that game has no model yet.
Aim methods and mouse out
Same features. You pick how long.
1 day, 3 days, week, month, quarter, six months, or a year. Activate the key in the desktop app. Start stays blocked until the key and a model are ready.
Get license