AI Topic (AI_TOPIC)

Answers questions from the documents of one of your topics. Use it as a tool an Agent can ask, or as a step in the flow.

Category: Integration (button AI Topic at the bottom of the palette) · Deployed: yes · Tool-capable: yes

Ports

Port Kind Purpose

Query (in_query)

Input

The question for the topic, when the module is a step

Answer (out_answer)

Output

The topic’s answer, when the module is a step

Tool (tool_mcp)

Tool

Wire to an Agent’s Tools port. The agent then asks the topic when it needs it

Properties

Field Required Meaning

Topic (topicId)

Yes

The topic whose documents are searched. The list shows RAG topics and GraphRAG topics; a GraphRAG topic carries "(GraphRAG)" after its name

Tool Filter (toolFilter)

No

Which functions of the topic the agent may use, by exact name. Empty: all of them. A topic offers one function, named after the topic, so leave this empty. A name that does not match hides the topic from the agent

How to add it: click AI Topic at the bottom of the palette. The dialog Select AI Topic shows every topic with its type (RAG or GraphRAG) and status. You can pick several and add them in one step. Each module is named after its topic.

Choosing another topic in the Topic field renames the module to the new topic’s name and clears the Tool Filter. Open Topic opens the selected topic’s page.

What happens at run time

  • As a tool: the agent sees the topic’s name and description. It decides itself when to ask the topic. A clear topic description helps it choose the right moments.

  • As a step: the previous module’s text goes to the topic as the question. The topic’s answer goes to the next module.

  • The topic answers with its own settings. Its models, retrieval settings and prompts all apply. The module only asks the question.

  • The topic checks the user’s access. The agent passes on the access of the user who chats. So the user needs access to the topic (its roles) and the chat permission for topics (for a GraphRAG topic: the GraphRAG chat permission). A scheduled run acts as the person who last saved the pipeline, with their access (see Scheduler).

  • No access does not break the answer. When the topic refuses the user, it is left out. The answer ends with a note that names the topic. As a tool, the note appears only when the question needed that topic. As a step, the note text becomes the step’s result.

  • Sources and pictures stay linked. Sources the topic cites appear under the agent’s answer in the agent chat. They open through the topic that cited them. Pictures the topic sends are shown too.

When the pipeline is checked

The module is complete when a topic is selected. The app reports it as missing, and refuses to assign the pipeline to an agent, when:

  • no topic is selected,

  • the selected topic was deleted, or

  • with Docker, the topic has no host port yet.

The address of the topic comes from the app itself. It works with Docker and with Kubernetes.

Use cases

  • Give an agent your documents without building retrieval in the pipeline.

  • Several knowledge bases in one agent: wire an HR topic and a product topic into the same Agent. The agent picks the right one for each question.

  • As a step, when the topic must always be asked: Chat Input → AI Topic → Agent, where the Agent reworks the topic’s answer.

  • Internal and external knowledge together: pair it with Web Search on the same Agent.

Notes

  • The topic must be running. If the agent cannot reach it, the run ends with an error. This is not checked when you save or assign the pipeline.

  • Access is checked against each user. When you pick a topic, or change the pipeline’s roles, the designer warns if some users of the pipeline cannot use the topic. It is a warning, not a block. For those users the topic is left out, with a note.

  • GraphRAG topics are experimental here. They work the same way, but treat them as a preview.

  • Rename with care. The agent finds the topic’s function by the topic’s name. After you rename a topic, restart the topic, then send the pipeline to the agent again (or restart the agent).

  • Topic to topic: wiring the Tool port of one AI Topic module into another AI Topic module connects the two topics directly. The second topic can then ask the first one itself. The app creates this link in the topic right away (it appears on the topic’s MCP Interfaces tab). Removing the wire removes the link.

  • Export and import: the selected topic does not exist in another installation. Select the topic again after an import.