AIMKU

A browser-based 3D aim trainer modeled after Valorant's aiming feel. It runs entirely in the browser, matches Valorant's sensitivity 1:1, and syncs profiles and weekly leaderboards through a Cloudflare Worker.

Live: aimku.xyz

Fan project for training purposes. Not affiliated with or endorsed by Riot Games.

Demo

A short walkthrough: start a session, choose a mode, play a round, and exit. (watch on YouTube)

Features

• True 3D FPS camera using the Pointer Lock API.
• Sensitivity matched to Valorant via its 0.07°/count yaw constant, so flicks transfer 1:1.
• Five training modes: Micro Flicks, Wide Flicks, Reflex Pop, Target Switch (gridshot), and Headshot Precision.
• GLB weapon viewmodel (Colt Navy revolver) with recoil and muzzle flash, plus a procedural fallback.
• Procedural hit/miss audio through the Web Audio API.
• Live HUD: score, accuracy, average split time, hit/miss counts, round timer, and FPS.
• Hitmarkers, floating score popups, and split-based dynamic scoring.
• Adjustable crosshair (color, size) and target size.
• Per-browser profiles and a weekly leaderboard, with no account required.
• Server-side score verification to guard against tampering.
• English and Indonesian UI.
• Fullscreen play with a minimal HUD. Desktop only; touch devices are blocked.

Tech stack

• Frontend: React 18, Three.js, Tailwind CSS, Vite
• Backend: Cloudflare Workers, D1 (SQLite), R2
• Hosting: Vercel (frontend), Cloudflare (Worker and storage)

Architecture

React (Vite)  ──HTTPS──>  Cloudflare Worker  ──>  D1 (SQLite)   profiles, scores, donations
localStorage              (worker.js)           └─>  R2          landing background images

• The frontend is deployed on Vercel and talks to the Worker over HTTPS.
• Profiles, scores, and donations are stored in D1. Players are identified by a per-browser deviceId; there is no login.
• Local settings (sensitivity, crosshair, target size, language, selected mode) persist in localStorage.
• Score submission is protected by a short-lived, HMAC-signed session token that can be redeemed once, plus per-device and per-IP rate limiting.

Getting started

Prerequisites: Node.js 18+ and npm.

npm install
npm run dev          # http://localhost:5173

To point the frontend at a deployed Worker during development, create .env.local:

VITE_API_URL=https://<your-worker>.workers.dev

Production build:

npm run build
npm run preview

Backend setup

The Worker and database live in worker.js, schema.sql, and wrangler.toml.

# Create the D1 database, then copy the returned id into wrangler.toml
npx wrangler d1 create valorant-aim-trainer-db

# Apply the schema
npx wrangler d1 execute valorant-aim-trainer-db --file=./schema.sql

# Deploy
npx wrangler deploy

CORS is restricted to aimku.xyz, localhost, and *.vercel.app preview deployments.

API reference

Method	Endpoint	Description
GET	/api/profile	Fetch a profile and best scores by deviceId
POST	/api/profile	Create or update a profile
POST	/api/session/start	Issue a signed, single-use session token
POST	/api/score	Submit a round score; verified server-side
GET	/api/rank	Get a player's current leaderboard rank
GET	/api/leaderboard	Weekly top scores, filterable by mode
GET	/api/donations	Recent supporters shown on the landing page
POST	/api/saweria-webhook	Donation webhook (Saweria)
GET	/api/backgrounds	List landing-page background images
GET	/api/bg/:name	Serve a background image from R2

Acknowledgements

• 3D model: "1851 Colt Navy Revolver" by Steven Jurriaans (CC BY).
• Inspired by Valorant (Riot Games). Non-commercial fan project.
• Built with the help of Claude (Opus 4.8) by Anthropic.

Notes

For aim that matches Valorant 1:1, disable OS mouse acceleration ("Enhance pointer precision" on Windows).