AI Knowledge Fabric

AI Knowledge Fabric is a modular toolkit for building RAG, GraphRAG, and agent pipelines, with automatic container orchestration for Docker and Kubernetes.

Core Capabilities

  • Modular construction kit for RAG, GraphRAG, and agent pipelines

  • Automatic container orchestration for RAG, GraphRAG, and agents

  • Orchestration for Docker / Kubernetes

Integrations & Storage

  • Seamless integration with your existing stack

  • Vector database options: CYROCK.DB, Qdrant, Milvus, Weaviate — each Topic stores its data in a vector database you register

  • Graph database options (GraphRAG): CYROCK.DB, Neo4j, Memgraph, or any other database that speaks Cypher over Bolt

Multi-LLM Support

  • Connect to any model provider of your choice — from local, self-hosted setups (Ollama, LM Studio and other OpenAI-compatible servers) to cloud model providers such as OpenAI, Anthropic, and Google Gemini

  • Different models per task: for example a small model for routing and a strong model for answers

Connectors

  • REST API for chat and document upload on every Topic, GraphRAG Topic, and Agent

  • MCP (Model Context Protocol): Topics can be used as tools by other Topics, by Agents, and by external MCP clients

  • Management REST API to create, configure, start, and stop Topics, GraphRAG Topics, and Agents from scripts

  • Pipeline modules for SQL databases, HTTP APIs, web search, files, and external MCP servers

Vector Search

  • Semantic similarity search across your entire context base for contextually relevant results

  • Optional keyword search (BM25) next to it, to find exact terms such as product numbers

Management Container

  • Provides a graphical user interface for configuring orchestration

  • Stores the connections to the systems it uses, including:

    • AI model servers, with their API keys

    • Vector databases, with their credentials

    • Graph databases, with their credentials

    • Docling servers

  • Optional OIDC integration (single sign-on)

  • Provides authentication / authorization for access to:

    • Topics and GraphRAG Topics (also when used as MCP tools)

    • Agent servers and their pipelines

RAG Topic Server

  • Configuration of a RAG system via a graphical user interface

  • Multi-document format support:

    • Text, binary (Word, PDF, etc.) via Docling, and images described by a vision model

    • Structured data import (JSON, XML, YAML, TOML, CSV, TSV): flattened into text, or narrated by a language model

    • Pre-configured REST interface for uploading files to insert and update the RAG

  • Configurable chunking:

    • Text pieces with configurable size and overlap

    • Optional contextual enrichment: a language model adds context to each text piece

  • Configurable retrieval strategy:

    • Dense (meaning) or hybrid (meaning + BM25 keywords)

    • Optional reranking by a language model

    • Routing: questions that need no documents are answered directly

  • Pictures from documents can be shown in answers

  • File attachments in the chat

  • Configuration of system prompt and all internal prompts

  • Authorized inter-topic connection via MCP

GraphRAG Topic Server

  • Configuration of a GraphRAG system via a graphical user interface

  • Converts unstructured and structured documents into a knowledge graph (entities and relationships)

  • Automatic insertion into the graph database

  • Automatic updating of the graph database, including removal of a deleted document’s contribution

  • Automatic vectorization of text pieces and entities

  • Automatic community detection with community reports

  • Configurable retrieval strategy:

    • LOCAL (details around specific entities), GLOBAL (overview across the whole collection), and HYBRID (both combined)

    • PATH (how two things are connected), plus direct answers and answers from chat attachments

    • A router picks the strategy for each question automatically

    • Dense or dense + BM25 keyword search, separately for LOCAL and GLOBAL

  • Configuration of system prompt and all internal prompts

Agent / Pipeline Server

  • Graphical pipeline designer with components including:

    • Chat input / output

    • LLM agent

    • AI Topic (RAG and GraphRAG Topics as tools)

    • MCP tools (including an attachment tool for chat files)

    • SQL DB

    • Key-value storage (read / write)

    • Structured output converter

    • API call

    • Web search

    • Python script

    • Data faker

    • File read and write

    • Embedding

    • Iterator

    • Loop

    • HITL (human-in-the-loop approval)

    • Scheduler

    • Guardrails

    • Validator

    • Text splitter and text format

    • Templates

    • …​and more

  • Multi-pipeline agent support

  • Support for pipelines and sub-pipelines

  • Support for Topics within a pipeline

  • Support for Graph Topics within a pipeline

  • Support for MCP within a pipeline

  • Import / export of pipelines

Deployment

  • Multi-cloud support: AWS, Azure, Google Cloud, or any Kubernetes environment

  • Fully containerized: complete Docker-based deployment model for consistency across all environments

  • Run-anywhere flexibility: from local development to on-premise data centers to enterprise cloud deployments

  • Single-command deploy: launch your entire agent infrastructure with docker compose in seconds

Metrics & Logging

  • Metrics dashboard – requests, response times, and token usage per model, for Topics, GraphRAG Topics, and pipelines

  • Live logs – the log of every Topic, GraphRAG Topic, and Agent container, viewable and filterable in the web UI

  • DEV reports – detailed step-by-step reports of single uploads, chat answers, and pipeline runs, for finding out why an answer looks the way it does

Security

  • Access control – role-based permissions, plus access rights per Topic, GraphRAG Topic, Agent, and Pipeline

  • Single sign-on – OIDC login with role mapping from your identity provider (multi-factor authentication is handled by your identity provider)

  • API tokens – personal access tokens for the management API, revocable at any time

  • Data sovereignty – keep your data in your chosen region or on-premise for regulatory compliance