AI Workflow Generation

Users describe what they want in plain language and the platform generates a runnable workflow through a multi-phase pipeline built for correctness — not a single prompt.

In this guide

Generation pipeline

• Natural-language-to-workflow planner with multi-step building and conversation checkpointing.
• A node-knowledge base gives the generator accurate metadata about every node type, so nodes are selected and configured correctly.
• Phased compile (add nodes → connect → parameterize), plan linting, and mid-phase validation that applies safe repairs automatically.
• Token-budget tracking and model routing — cheap, fast models classify while stronger models reason.
• Graceful fallback to a legacy path if a build phase cannot complete, so the user still gets a result.

Prompt Gallery — per-model prompt tuning

• A structured library of generation prompts tuned per model, so the same pipeline runs on different foundation models at their best without any code change.
• Each pipeline stage (intent, supervisor, discovery, planner, builder, parameter-updater, responder) reads a prompt resolved exact-model → model-family → default.
• A model only needs override files for the stages that differ, alongside a complete default set.

AI Models management

• Create, list, discover, update, and delete AI models; set a platform default; multi-provider support.