A one-step pipeline is 5 lines of YAML. No boilerplate, no classes, no setup code.
Composable fleets
Pipelines are steps inside other pipelines. Build small automations, assemble them into large systems.
Provider-agnostic
Swap Claude for GPT for Llama without changing your pipeline. One spec, any model.
Built for frontier models
Designed and tested with Anthropic Claude 4.x and OpenAI GPT-5 Codex models. These handle multi-step agent workflows reliably — results with smaller or older models are unpredictable.
Runs locally
Your pipelines execute on your own machine or infrastructure. No data leaves your network unless you choose a hosted model provider.