winget install --id=saransh121.flowChart -e Describe a process or a topic and flowChart turns it into an editable mind map or flowchart. A small open model (Qwen3-1.7B) runs on your own CPU or GPU, so nothing leaves your machine. Nodes stream onto the canvas while generating; rename, reshape, connect and export to PNG, PDF, PowerPoint or Word. Optionally bring your own Anthropic, OpenAI, Gemini or OpenRouter API key for richer diagrams.
Type what you mean. Get an editable mind map or flowchart. No internet, no account, no data leaves your machine.
Demo video — see docs/demo.mp4 in the repo if it doesn't play here.
Package managers (after the manifests land):
winget install saransh121.flowChart # Windows
brew install --cask saransh121/tap/flowchart # macOS
Grab the latest build for your OS from the Releases page:
| OS | File |
|---|---|
| Windows | flowChart--win-x64.exe (installer) or .zip (portable) |
| macOS | flowChart--mac-arm64.dmg (Apple Silicon) or -mac-x64.dmg (Intel) |
| Linux | flowChart--linux-x86_64.AppImage or -linux-amd64.deb |
On first launch the app downloads the model once (1.1 GB) into its data folder. After that it never needs the network again.
> Windows SmartScreen / macOS Gatekeeper: builds are not code-signed yet. On Windows click More info → Run anyway. On macOS right-click the app → Open, or run xattr -dr com.apple.quarantine /Applications/flowChart.app.
Drop any llama.cpp .gguf file into the models folder and restart. The largest file wins.
%APPDATA%\flowChart\models~/Library/Application Support/flowChart/models~/.config/flowChart/modelsQwen3 0.6B (400 MB) is 2.5x faster and fine for simple diagrams. Anything that chats works, but models under 1B tend to misclassify processes as mind maps.
| Machine | Speed |
|---|---|
| Laptop with integrated GPU (Vulkan), 8 GB RAM | ~15 tok/s, a 12-node flowchart in ~20 s |
| Same laptop, 0.6B model | ~35 tok/s |
| NVIDIA / AMD dedicated GPU | much faster, auto-detected |
CPU-only with 8 GB RAM and the 1.7B model is slow. Use the 0.6B model there.
See docs/LAUNCH.md for the winget / Homebrew / store submission steps and ready-to-paste launch posts.
npm install
npm run dev # Vite + Electron with hot reload
npm run pack # unpacked build in release/
npm run dist # installers for the current OS
npm run spike # the model benchmark used to pick the default model
Put a .gguf in ./models/ for development, or let the app download one.
{type, title, nodes, edges}. It never emits coordinates.MIT. The default model is Qwen3-1.7B by Alibaba, Apache-2.0, downloaded from Hugging Face on first run.