Concepts¶
This section explains the core ideas behind Pipelit. Understanding these concepts will help you design effective workflows and get the most out of the platform.
Workflows¶
The central organizing unit in Pipelit. A workflow is a visual pipeline you design on a canvas, connecting triggers, agents, tools, and logic nodes into an executable graph.
Nodes & Edges¶
Nodes are the building blocks of workflows. Edges are the connections between them. Together they form a typed, directed graph with validated data flow.
Triggers¶
Triggers are specialized nodes that initiate workflow execution. Pipelit supports chat, Telegram, manual, scheduled, workflow, and error triggers -- all as first-class nodes on the canvas.
Agents¶
LLM-powered nodes that reason, call tools, and produce responses. Agents use LangGraph's ReAct architecture and support conversation memory, tool calling, and model selection.
Tools¶
Sub-component nodes that give agents capabilities: execute shell commands, make HTTP requests, search the web, evaluate math, and check the current time.
Expressions¶
Jinja2 template expressions let you reference upstream node outputs in system prompts and configuration fields. Use {{ nodeId.portName }} syntax with filters and fallbacks.
Execution¶
How workflows run: trigger-scoped compilation, topological node ordering, real-time WebSocket status updates, and result propagation through the graph.
Memory¶
Persistent knowledge storage across executions. Conversation memory gives agents continuity. Global memory stores facts, episodes, and procedures that any agent can recall.
Epics & Tasks¶
A task delegation system for multi-agent coordination. Epics group related tasks with budgets and deadlines. Agents can create, query, and update them autonomously.
Cost Tracking¶
Per-execution token counting and USD cost calculation. Tracks input/output tokens across all LLM calls with Epic-level budget enforcement.