Build local AI workflows and inspect what actually ran.
Pantograph is a Rust workflow framework with a Svelte workbench and an optional Tauri desktop host. Connect processing and inference nodes, let the scheduler manage execution, and inspect task status, diagnostic events and retained outputs. Applications can also use the headless workflow service.
Actual browser workbench with untrained CPU fixtures. Demo details.
Use the pinned Rust 1.92.0 toolchain, Node.js 24.12.0, npm 11.6.2 and Python 3. From a checkout:
git clone https://github.com/MrScripty/Pantograph.git
cd Pantograph
npm install
python3 examples/scheduler-demo/serve.pyOpen http://127.0.0.1:8765. Run the supplied JSON, then click Test expected rejection and Run workflow to see a real scheduler rejection.
For the full workbench, run npm run dev in another terminal and follow Open the live workbench. On Graph, load editor-demo, edit Text Input, Save, then Submit. This uses real Rust execution and temporary demo state. Demo guide.
To reproduce the embedding workflow pictured above, follow the CPU embedding host guide. It uses tiny committed numerical fixtures, with no model download or paid inference.
Install your platform's Tauri prerequisites using the development guide, then:
npm run dev:desktop
Use npm run build:desktop to package it. The browser demo and native desktop host have different capabilities; the screenshot above does not qualify a native Tauri window. The separate Node Lab authoring surface remains incomplete.
- Architecture and Rust crate map
- Development and checks
- Runtime setup
- Headless integration
- Research and documentation index
The repository includes Apache-2.0 license material while Cargo metadata declares MIT OR Apache-2.0. Resolve that mismatch before distribution; see release guidance.