Components¶

Pipelit provides 42 component types that serve as the building blocks for workflows. Each component type represents a distinct piece of functionality -- from receiving events to reasoning with LLMs to routing data through conditional logic.

Components are organized into seven categories, each with a distinct role in the workflow pipeline.

How Components Work¶

Every component on the canvas is a node in the workflow graph. Nodes connect to each other via edges that carry typed data between output ports and input ports.

Port System¶

Each component type defines its input ports and output ports with specific data types:

When you draw an edge between two nodes, Pipelit validates that the source output port type is compatible with the target input port type. Incompatible connections are rejected with a 422 error.

Execution Model¶

Components follow a consistent execution pattern:

1. The orchestrator resolves Jinja2 expressions in the component's configuration (system prompt, extra config).
2. The component's run() function executes with the current workflow state.
3. The component returns a flat dict with port values (e.g., {"output": "result text"}).
4. The orchestrator wraps non-underscore keys into node_outputs[node_id] for downstream access.
5. A node_status WebSocket event is published with the result status.

Sub-Component Connections¶

AI nodes connect to sub-components via special diamond-shaped handles at the bottom of the node:

Accessing Upstream Data¶

Use Jinja2 template expressions to reference data from upstream nodes:

{{ trigger.text }}
{{ nodeId.portName }}
{{ agent_abc123.output | upper }}

The trigger shorthand always refers to whichever trigger fired the current execution.