FlowSpec was born out of my frustration with ad-hoc “scripts everywhere” setups. I wanted something that makes AI integration both standardized and composable—so I put together a schema and a simple toolchain for describing and coordinating AI tasks. It’s heavily inspired by container-based CI/CD tools, but specifically for orchestrating AI in production.

Right now, FlowSpec describes automation flows in simple steps, as well as the transformations or sanity checks in between. I’m still experimenting with how to keep it flexible yet opinionated enough to encourage best practices. I’d love your feedback on that balance!

I’m also thinking about new ways to make FlowSpec more interactive and transparent, possibly through a web UI that visualizes the DAG of AI tasks. I believe part of what makes an AI workflow robust is clarity on all the steps—and ideally, a way to replay or tweak them as your models evolve. If you have any ideas on features or improvements, please let me know. I’m eager to see how other developers structure their AI pipelines.

Appreciate your feedback, here or via issue/PR. This will evolve as my challenge plays out, but I know there are much smarter automators here who I'd love to hear from.

Links

• Source code + examples: https://github.com/woodyhayday/FlowSpec

• Intro & explainers: https://profitswarm.ai/flowspec-workflow-schema/