Module Reference

One page per module type of the Pipeline Designer. Each page lists the module’s ports, its required and optional properties, whether it can be used as a tool, and what it is for. There are 29 module types. 28 of them are in the palette. The Attachment Tool is not (see below).

How to read these pages

Heading Meaning

Heading	Meaning
Category	Where the module sits in the palette
Deployed	Whether the module is sent to the agent at all. `No` means it exists only on the canvas
Tool-capable	Whether it can be wired into an Agent’s Tools port and called by the model
Ports	`Input` receives the output of the step before · `Output` passes on to the next step · `Tool` connects to an Agent’s Tools port
Required	A required property that is empty makes the pipeline impossible to deploy

Category

Where the module sits in the palette

Deployed

Whether the module is sent to the agent at all. No means it exists only on the canvas

Tool-capable

Whether it can be wired into an Agent’s Tools port and called by the model

Ports

Input receives the output of the step before · Output passes on to the next step · Tool connects to an Agent’s Tools port

Required

A required property that is empty makes the pipeline impossible to deploy

Check the required properties first. A module with an empty required property cannot be deployed. The app refuses to assign such a pipeline to an agent and names the empty fields. A pipeline that is already on a running agent can still be saved, but the new version is not sent to the agent. A message names the empty fields. If a change does not show up, check the required properties first. See Pipeline Editor.

Palette

Input & Output

Chat Input — entry point, receives the user’s message
Chat Output — exit point, returns the result

Models & Agents

Agent — the reasoning step; the only module with a Tools input
Embedding — turns text into a vector

Flow Control

Scheduler — starts the pipeline on a schedule (cron)
Iterator — walks through a JSON list, one item at a time
Loop — repeats steps until a Validator passes
Validator — checks a value with a condition, a script or a model, with Pass and Fail
Guardrails — safety check with Pass and Fail

Data Sources

API Call — HTTP request
SQL Database — SQL query
Data Faker — made-up test data
KV Store Read — reads a stored value by key (Docker only)
KV Store Write — stores a value under a key (Docker only)
Structured Output — a model turns text into a JSON list

Files

File Read — reads a file from the agent’s file area, S3 or Azure Blob
File Write — writes a file to the agent’s file area, S3 or Azure Blob

Utilities

Template — fills placeholders in a text
Text Splitter — cuts a text into pieces
Text Format — one text operation
Python Script — runs a Python script in its own container
Web Search — web search, used as a tool
Description — a note on the canvas, never deployed

Human Interaction

Human Approval — pauses the run until a person answers

Integration (buttons below the palette)

MCP Tool — a tool from the catalog, run by the agent
SSE Tool — an existing MCP server reached over HTTP/SSE
AI Topic — a RAG Topic or GraphRAG Topic as a tool or step
Pipeline — another pipeline as a tool or step

Not in the palette

Attachment Tool — lets an agent read a file a user attached in the chat. The palette has no entry for it, so it is only available in pipelines that already contain one, for example imported ones

Tool-capable modules

These 14 modules can be wired into an Agent’s Tools port. Anything else wired there is ignored:

MCP Tool · SSE Tool · Attachment Tool · API Call · AI Topic · SQL Database · Data Faker · KV Store Read · KV Store Write · File Read · File Write · Pipeline · Web Search · Python Script

Not tool-capable are Human Approval (a paused run cannot hand control back to the model’s tool loop), Embedding, Template, Text Splitter, Text Format, Structured Output, and all flow-control modules.

Step or tool

Many modules work either way, and the difference matters:

	As a step	As a tool
Wired to	Input and Output ports	An Agent’s Tools port
Runs	Always, in order	Only when the model decides to call it
Input	The output of the step before	Arguments the model chooses
Description field	Optional	Important — it is how the model knows when to call it

As a step

As a tool

Wired to

Input and Output ports

An Agent’s Tools port

Runs

Always, in order

Only when the model decides to call it

Input

The output of the step before

Arguments the model chooses

Description field

Optional

Important — it is how the model knows when to call it

Docker and Kubernetes

Some modules keep data on the machine of the agent. This works with Docker, but not on Kubernetes:

Module	Docker	Kubernetes
KV Store Read / Write	Values are kept on a volume and survive stop and start	The agent refuses the pipeline
File Read / Write with the local file area	The files are kept only while the agent container exists	The agent refuses the pipeline
Python Script	Runs in a short-lived container	Runs as a Kubernetes job
MCP Tool (Docker based)	Runs in its own container	Runs in its own pod

Module

Docker

Kubernetes

KV Store Read / Write

Values are kept on a volume and survive stop and start

The agent refuses the pipeline

File Read / Write with the local file area

The files are kept only while the agent container exists

The agent refuses the pipeline

Python Script

Runs in a short-lived container

Runs as a Kubernetes job

MCP Tool (Docker based)

Runs in its own container

Runs in its own pod

For files on Kubernetes, use S3 or Azure Blob storage instead. See the pages of the modules and Installing on Kubernetes.