AI Knowledge Fabric — User Guide

What is AI Knowledge Fabric?

AI Knowledge Fabric is a platform for building, running, and managing AI-powered services — without writing code. You configure AI assistants through a web interface, and the platform automatically provisions the underlying infrastructure for you.

The platform covers four main areas: RAG Chatbots (Topics), GraphRAG Topics, Agents, and Pipelines. Each serves a different purpose and can be used independently or combined.

Want to build something right away? Getting Started walks the full path from installation to your first grounded answer, in order, with links to the detail pages.

RAG Chatbots — Topics

A Topic is a private AI chatbot that answers questions based on your own documents. "RAG" stands for Retrieval-Augmented Generation — the AI searches your uploaded content first, then uses that knowledge to produce an accurate, grounded answer instead of relying solely on its general training.

What you can do with a Topic:

  • Upload your own documents — PDFs, Word files, text files, and more. The platform indexes them automatically so the chatbot can search them.

  • Ask questions in natural language — The chatbot retrieves the most relevant passages from your documents and uses them to answer.

  • Run multiple Topics in parallel — Each Topic is completely isolated. You can have one for HR policies, another for technical manuals, another for customer FAQs.

  • Connect external tools via MCP — Extend a chatbot’s capabilities by linking it to other running Topics or external tool servers.

  • Customise the chatbot’s behaviour — Each Topic has its own system prompt so you can give it a specific role, tone, or set of instructions.

Typical workflow:

  1. Create a Topic with a name and description.

  2. Choose the language model and embedding model to use.

  3. Start the Topic — the platform spins up the required services automatically.

  4. Upload your documents on the Data Upload tab.

  5. Click the Topic name in the list to open the chat interface.

GraphRAG Topics

A GraphRAG Topic is the graph-based sibling of a Topic. Next to the text pieces, it builds a knowledge graph from your documents: the things they mention (entities) and how they are connected. It groups the graph into topical communities and writes a short summary of each one.

What it adds to a Topic:

  • Questions about connections — "How is A related to B?" is answered by following the graph.

  • Overview questions — "What are the main topics?" is answered from the community summaries.

  • Automatic choice — for every question, a router picks the best way to search.

It needs a registered graph database in addition to the vector database, and uploads take longer, because a model reads every text piece. See GraphRAG — Getting Started.

AI Agents

An Agent is an autonomous AI assistant that can reason, plan, and take actions across multiple steps. Unlike a simple chatbot that answers a single question, an Agent can break down a task, use tools, call other services, and work through a multi-step process to reach a result.

Agents run one or more Pipelines (see below) that define how the Agent thinks and what tools it has access to.

What you can do with an Agent:

  • Run complex, multi-step tasks — The Agent decides on its own which steps to take to fulfil a request.

  • Connect tools — Agents can use MCP tool servers to retrieve information, call APIs, or trigger actions in external systems.

  • Run multiple independent Agents — Each Agent runs in its own isolated container on a dedicated port.

  • Select which Pipeline to use per conversation — A single Agent can run several different Pipelines; you choose the one that fits your task when you open the chat.

Typical workflow:

  1. Design a Pipeline that defines the Agent’s instructions, model, and tools.

  2. Create an Agent and assign the Pipeline to it.

  3. Start the Agent — the platform starts the container and registers all assigned Pipelines automatically.

  4. Open the Agent chat, select a Pipeline, and start a conversation.

Pipelines — Visual Workflow Designer

A Pipeline defines how an AI Agent works: which language model it uses, what instructions it follows, and which tools it can call. Pipelines are built visually in the Pipeline Designer — a drag-and-drop canvas where you connect nodes together.

What you can do in the Pipeline Designer:

  • Add Agent nodes — Each Agent node represents one AI step. You configure its model, provider, and system instruction directly in the properties panel.

  • Connect tools — Drag an MCP Tool node onto the canvas and connect it to an Agent node to give that Agent access to an external tool server.

  • Embed other Pipelines — Use a Pipeline node to call another Pipeline as a reusable sub-step.

  • Link to a Topic — Use an AI Topic node to route part of a conversation to one of your running RAG chatbots.

  • Define entry and exit points — A Chat Input node receives the user’s message; a Chat Output node returns the final response.

  • Export and import — Pipelines can be exported as JSON and imported on another instance.

Node types at a glance:

Node Purpose

Agent

An AI step with its own model, instructions, and optional tools

MCP Tool

An external tool server the Agent can call

AI Topic

A running RAG Topic used as a knowledge source

Pipeline

Embeds another Pipeline as a reusable sub-step

Chat Input

Entry point — receives the user’s message

Chat Output

Exit point — returns the final response

MCP Tools

MCP (Model Context Protocol) is a standard way to connect AI services to external tools and data sources. The platform includes a built-in MCP Tool Browser where you can discover available tool servers from a public catalog and connect them to your Topics or Agents.

What you can do:

  • Browse the MCP tool catalog to find servers for web search, file access, APIs, and more.

  • Connect MCP tool servers to a running Topic so the chatbot can call them during a conversation.

  • Add MCP Tool nodes in the Pipeline Designer to give an Agent access to external tools.

Services Summary

Area What it provides

Topics (RAG Chatbots)

Document-grounded Q&A chatbots backed by your own uploaded content

Agents

Autonomous AI assistants that execute multi-step tasks using tools and Pipelines

Pipelines

Visual workflow definitions that configure how an Agent reasons and which tools it uses

MCP Tools

Connections to external tool servers that extend what Topics and Agents can do

GraphRAG Topics

Q&A chatbots that also build a knowledge graph, for questions about connections and overviews

AI Servers (admin)

Managed Ollama inference servers and external model provider connections

Docling Servers (admin)

Document pre-processing servers used to convert PDFs/Office files into clean text before embedding

Logs

Every running Topic, GraphRAG Topic and Agent has its own Logs tab with a live log viewer: it shows recent log history and then streams new entries as they happen, so you can watch a chatbot or agent’s activity in real time without shelling into the container.

Roles & Permissions

Access is controlled by roles built from fine-grained permissions (e.g. being able to view Topics without being able to delete them), not a single "admin vs. everyone else" switch. What you see in the navigation menu — and which actions are available on each screen — depends on the permissions your role grants. If something you expect to see is missing, ask an administrator to check your assigned role under Admin → Users.

Administrators can also enable Single Sign-On (OIDC, e.g. via Keycloak or Entra ID) instead of, or alongside, local username/password login — ask your administrator which login method applies to your organisation.