FlowDrop Workflow Orchestration
FlowDrop is an open-source workflow orchestration engine with a visual editor. A workflow is a graph of typed nodes with explicit edges, and every run can be traced back to the steps it took.
It is useful where an AI feature is a pipeline rather than a single model call: load context, build a prompt, call a model, validate and map the answer. Narrow, deterministic steps keep language models on a short leash. FlowDrop runs inside Drupal, connected to its content, users and permissions. A standalone engine written in Rust is in development and currently in a closed beta without public access. A published specification describes the behaviour both implementations follow. The Drupal distribution Atelier is built on it.
Updates
FlowDrop is developed by our team as open source. On a publisher's editorial website it runs in production: workflows suggest tags and categories for articles, picking only from the existing vocabularies. We also propose it as the orchestration layer for AI features in other Drupal projects, and we maintain an open benchmark that compares workflow designs on correctness, cost and speed. We are also building a standalone FlowDrop engine in Rust; it is in a closed beta and not publicly available yet. It is in production on one site so far, so it sits in trial.