AI Components¶

AI components are the LLM-powered nodes in Pipelit. They send messages to a language model, interpret the response, and produce structured outputs that drive the rest of your workflow.

Overview¶

There are five AI component types:

All five share a common trait: they require an AI Model sub-component connection to function. Without a model, the node cannot resolve which LLM to use and will fail at build time.

Canvas Appearance¶

AI nodes have a distinctive visual treatment on the workflow canvas:

• Fixed 250px width with a separator line dividing the node header from the sub-component pills below.
• Bottom diamond handles for connecting sub-components (model, tools, memory, output parser).
• Left circle handle for the input connection (messages).
• Right circle handle for the output connection.

Sub-Component Support¶

Each AI node type supports a different set of sub-components. Connect these via the colored diamond handles at the bottom of the node.

Handle colors:

Common Input¶

All five AI nodes accept the same input:

The messages input typically comes from a trigger node (Chat, Telegram, etc.) or from an upstream node that produces LangChain messages. The messages list carries the full conversation context that the LLM uses to generate its response.

System Prompt and Jinja2¶

All AI nodes support a system prompt that instructs the LLM on its behavior. System prompts support Jinja2 template expressions, resolved by the orchestrator before execution:

You are a {{ trigger.payload.role }} assistant.
The user said: {{ trigger.text }}
Previous output: {{ code_abc123.output }}

See Expressions for the full template syntax.

What's Next?¶

• Agent -- autonomous reasoning with tool calling
• Deep Agent -- advanced agent with built-in task planning, filesystem tools, and subagents
• Categorizer -- classify input into categories
• Router -- route execution based on input content
• Extractor -- extract structured data from text