Gradio's WorkflowCanvas 0.4.0 lets you run subgraphs as API endpoints

Gradio, the go-to Python library for quickly building UIs around machine learning models, just shipped version 0.4.0 of its WorkflowCanvas component. This release introduces a major capability: running subgraphs as independent API endpoints. Each subgraph defined inside a Gradio workflow is now automatically exposed as a named endpoint, reusing the existing /info, /call, and /api routes. This means developers can treat complex pipelines as composable microservices, calling any part of a workflow directly from code or external tools. A new 'View API' panel inside the canvas gives immediate visibility into each subgraph's inputs, outputs, and callable URL — streamlining debugging and integration. The update also refactors the space and model discovery modals, making it easier to find and connect to remote resources. Contributors abidlabs and hannahblair drove this release, which includes dependency bumps for @gradio/atoms, statustracker, utils, and client.

For teams building multi-step AI pipelines — from data preprocessing to model inference to post-processing — this feature dramatically reduces boilerplate. Instead of manually wiring subgraph invocations, developers can now treat each step as a first-class API endpoint. Combined with Gradio’s existing real-time streaming and UI generation, WorkflowCanvas 0.4.0 makes it possible to create modular, testable, and shareable AI workflows without leaving the Python ecosystem. The shift toward subgraph-as-API also opens the door to more advanced patterns like multi-agent orchestration and conditional branching, all managed through Gradio’s familiar interface. With over 43k GitHub stars and a vibrant community, this release solidifies Gradio’s role as a key tool for productionizing AI demos and internal tools.