Welcome to EdgeWeave

EdgeWeave is a desktop app for building data-science and AI workflows as graphs. You drag nodes onto a canvas, wire them together and press Run. Each node shows its result as it finishes, and the whole graph exports to a plain Python script or a web dashboard when you're done.

It ships with nodes for data loading and cleaning, Plotly charts and maps, scikit-learn, PyTorch, NetworkX graphs, SQLite, 3D viewers, simulation (SimWeave), language models and retrieval (RAG), and more. PostgreSQL and a statistics toolkit (tests, copulas) are free modules in the Marketplace. You can write your own nodes in a few lines of Python.

Start here

• Getting Started: install, run a demo, build and export your first workflow.
• Features overview: a map of everything the app does and where to find it.
• Projects and workflow files and Python packages and environments: where your work lives and how to add libraries.

Tutorials

1. Build a graph with the Ctrl+K palette
2. Version workflows with Git and see visual graph diffs
3. Export a dashboard
4. Chat with the assistant and run the RAG demo
5. Export a workflow to Python
6. Run workflows on a schedule
7. Install modules from the Marketplace
8. Write your own node with @node

Reference

• Subsystems: call one .weave file from another.
• Control flow: for-loops on the canvas.
• Troubleshooting: fixes for the most common problems.
• Node reference (online): every shipped node, with its ports, fields and Python-export status.
• Node families: AI / LLM and RAG, SimWeave continuous, SimWeave discrete.

Inside the app, right-click any node for its own documentation, including its ports and whether it supports Python export.

For contributors

Layer	Tech	Role
Shell	Electron	Desktop window, file dialogs, installer
Frontend	Drawflow + CodeMirror + Vite	Graph canvas, code editor, sidebar, settings
Backend	FastAPI (Python 3.11)	Node registry, graph parsing, REST + WebSocket
Execution	Worker subprocess	Runs the parsed DAG, streams progress
SDK	sdk/node.py, sdk/decorators.py	How nodes are authored and registered

The on-disk graph format is .weave: JSON produced by Drawflow's export() plus a small EdgeWeave wrapper.

• Shipped nodes live in backend/core/nodes/ (@register_block) and must pass the metadata audit (backend/tests/test_codegen_audit.py); user and marketplace nodes use @node (sdk/node.py).
• Path constants live in backend/core/config.py; never hard-code paths elsewhere.
• Workflow-state mutations must take _workflow_lock (backend/utils/state.py).
• Don't add node-type dispatch to backend/utils/parse_drawflow.py; that parser is intentionally generic.
• Design notes and investigation logs are under Contributor notes in the sidebar.