In a blog post published August 25, 2026, Hugging Face introduced gr.Workflow, a feature built into its Gradio library that lets developers describe AI pipelines as a graph of typed nodes. According to the post, authored by Yuvraj Sharma and Abubakar Abid, the tool serves a drag-and-drop canvas where every node is runnable and every intermediate result is visible; the same graph also functions as a REST API and can be deployed to Hugging Face Spaces with one command. The post describes three node types: references (inputs), operators (steps that do work), and subjects (outputs). An operator can be a user's own Python function, a model on Hugging Face Inference Providers, another Gradio Space, or a row from a Hub dataset. Hugging Face provides live example Spaces, including an image editor calling Qwen-Image-Edit, a media studio chaining FLUX with background removal and text-to-speech, a parallel 'fan-out' image generator, and a dataset profiler using the Datasets Server API. The post notes that fn nodes can run models on a GPU inside a Space using the @spaces.GPU decorator with ZeroGPU. Each workflow output automatically becomes a REST endpoint callable via the Gradio client or curl. Hugging Face says a follow-up post will cover building a more involved app.
- gr.Workflow was announced by Hugging Face on August 25, 2026
- Workflows use three node types: references, operators, and subjects
- Each workflow output automatically becomes a callable REST endpoint
- Nodes can call Inference Providers, other Gradio Spaces, Hub datasets, or run models on GPU via ZeroGPU
What it means for you
If you or your team build AI apps that string several steps together — generate an image, then edit it, then caption it — this new Gradio feature lets you lay those steps out visually and see the result of each one, instead of debugging Python line by line. It also turns your pipeline into a shareable web app and an API automatically. This is a developer tool, so it matters only if someone on your side actually writes code.
Try this
Open one of the live example Spaces linked in the post (such as the image editor), click 'Duplicate,' and rewire it to test whether this visual approach fits how your team builds.
Who should care
Developers and technical small-business teams who already prototype AI features with Gradio or Python and want faster ways to chain and deploy multi-step pipelines.
Skip this if
No one on your team writes code, or you rely on off-the-shelf AI apps rather than building your own pipelines.
Sources: Hugging Face — read the original