Desktop App · Windows
FPSOpti AI
A local AI-powered PC performance optimizer. Scan your hardware, pick a game and target FPS, and get optimized graphics settings generated by a small LLM running entirely on your machine.
Windows 10 / 11 · ~ a few hundred MB once Ollama + the model are installed
What it does
One-click PC scan
Auto-detects CPU, RAM, GPU, VRAM, driver version, storage, and display resolution.
Local AI report
Streams a per-setting recommendation from a qwen3:0.6b model running through Ollama.
Fully on-device
No telemetry, no remote API. Specs and prompts never leave your PC.
Export to .txt
Save the AI report locally and revisit it any time.
Quick Setup
How to install
FPSOpti AI uses a model running locally through Ollama, so the install is three short steps.
Install Ollama
Download and run the installer from ollama.com/download. After install it runs in the background and exposes a local API on http://localhost:11434.
Pull the AI model
Open PowerShell and run this once. It downloads the small model FPSOpti AI uses (~ a few hundred MB).
ollama pull qwen3:0.6bRun FPSOpti AI
Double-click FPSOptiAI.exe. No Python or extra setup required. The first launch can take a couple of seconds while the bundled assets unpack.
How to use it
- 1
Click Scan My PC. Your hardware fills in the Detected Specs card.
- 2
Optionally edit any spec the scanner got wrong (CPU, GPU, RAM, resolution).
- 3
Fill in the game form: name, target FPS, in-game resolution, priority (Max FPS / Balanced / Best Quality / Competitive), and any notes.
- 4
Click Generate Optimization Plan. The AI report streams in live.
- 5
Click Export as .txt to save the report.
100% local — no data leaves your PC
Hardware scans run on-device. The AI recommendation is generated by a model running locally through Ollama, so your specs, game info, and notes are never uploaded anywhere.
Troubleshooting
"Ollama does not seem to be running."
Open the Ollama app, or run ollama run qwen3:0.6b in PowerShell, then click Generate again.
"The model ‘qwen3:0.6b’ is not installed."
Run ollama pull qwen3:0.6b in PowerShell.
Generation is slow on the first run
The first request loads the model into memory. Subsequent requests stream much faster.
Some specs show as "Unknown"
Older Windows installs or virtualized GPUs don't always expose VRAM or driver version. Use the manual override fields under Detected Specs to fill them in.
Windows SmartScreen warning on first launch
Single-file PyInstaller builds occasionally trigger a SmartScreen prompt because the .exe isn't code-signed. Click More info → Run anyway if you trust the source (this repo).
Ready to tune your settings?
Grab the latest build and run it locally. Source is on GitHub if you want to inspect or modify it.