A workflow is a structured execution of logical steps, defined as a sequence of modular tasks which call chips, orchestrated within a declarative pipeline.

Each workflow represents a segment of intellectual labor - such as decision-making, analysis, or automation - broken down into reusable, well-defined operations.

These workflows are described in YAML, executed top-down unless redirected by flow controls (like branching or sidequests), and can span across multiple pipelines for complex or multi-phase reasoning.

By combining dynamic execution, asynchronous processing, and real-time feedback, workflows enable scalable, intelligent automation of thought-driven processes.

Pipeline

A pipeline is a declaratively defined sequence of tasks, written in YAML. It models a segment of intellectual work — such as a decision-making process, reasoning chain, or task automation flow. Pipelines are:

• Executed top-down unless flow is redirected
• Composed of well-defined tasks
• Parameterizable, observable, and interruptible

Task

A task is a step within a pipeline. It maps directly to a chip invocation and supports fields like:

• name: Human-readable identifier
• method_name: Fully qualified chip name
• params: Input arguments for the chip
• Optional: log, save_as, async_as, when

Chip

A Chip is the atomic unit of work in ReasonChip. It performs a single, well-defined action (e.g., sending a message, running a query, making an API call). Each chip is:

• An async Python function
• Accepts and returns Pydantic models
• Registered with the system via @Registry.register
• Discoverable and reusable within pipelines

ReasonChip pipelines support advanced control patterns:

• Branching: Jump to a new pipeline
• Sidequest: Run a sub-pipeline and return
• Restart / Stop / Cancel: Control execution flow programmatically