Skip to main content

​
Welcome to Mimir AIP

Mimir AIP is an ontology-driven platform for data aggregation, processing and analysis. It provides a unified runtime for data ingestion pipelines, machine learning model training and inference, and digital twin management — all backed by a persistent metadata store and exposed as Model Context Protocol (MCP) tools for direct use by AI agents and LLM-based workflows.

Quick Start

Get up and running with Docker Compose in minutes

Installation

Full installation guide for Docker Compose and Kubernetes

MCP Integration

Connect Mimir AIP to AI agents and LLM workflows

API Reference

Explore the REST API and MCP tools

​
Key Features

Build configurable pipelines that ingest data from multiple sources, process it through transformation steps, and output to various storage backends. All data is normalized to CIR (Common Internal Representation) format for consistency across backends including PostgreSQL, MySQL, MongoDB, S3, Redis, Elasticsearch, and Neo4j.
Train and deploy machine learning models constrained by your ontology. Supports decision trees, random forests, regression, and neural networks. Workers handle training and inference as Kubernetes jobs, enabling scalable ML workflows.
Create live, queryable in-memory graphs of your entities. Digital twins are initialized from OWL/Turtle ontologies and synchronized from storage backends. Query them using the built-in SPARQL engine.
All platform capabilities are exposed as 55 MCP tools, allowing AI agents to create projects, configure pipelines, train models, query digital twins, and more using natural language. Works with any MCP-compatible client including Claude Code.

​
Architecture Overview

Mimir AIP consists of three main components: 
┌──────────────────────────────────────────────────────┐
│                      Client Layer                    │
│   Web Frontend (port 3000)   │   MCP Client / Agent  │
└──────────────┬───────────────┴──────────┬────────────┘
               │  REST API                │  SSE (MCP)
               ▼                          ▼
┌─────────────────────────────────────────────────────┐
│                    Orchestrator                     │
│  ┌──────────┐  ┌──────────┐  ┌──────────────────┐  │
│  │ Projects │  │Pipelines │  │   ML Models      │  │
│  │ Ontology │  │Schedules │  │   Digital Twins  │  │
│  │ Storage  │  │  Queue   │  │   MCP Server     │  │
│  └──────────┘  └────┬─────┘  └──────────────────┘  │
│          SQLite     │                               │
└─────────────────────┼───────────────────────────────┘
                      │  Kubernetes Jobs
                      ▼
           ┌─────────────────────┐
           │       Workers       │
           │  (pipeline, train,  │
           │   infer, DT sync)   │
           └──────────┬──────────┘
                      │
          ┌───────────▼───────────┐
          │   Storage Backends    │
          │  Filesystem · Postgres│
          │  MySQL · MongoDB · S3 │
          │  Redis · ES · Neo4j   │
          └───────────────────────┘

Orchestrator

The long-running HTTP server managing all metadata in SQLite. Exposes REST API and MCP SSE endpoint. Spawns workers as Kubernetes jobs.

Worker

Short-lived Kubernetes jobs that execute pipelines, train ML models, run inference, and sync digital twins. Designed for horizontal scalability.

Frontend

React/TypeScript single-page application for user-friendly platform management. Communicates exclusively with the orchestrator REST API.

​
Core Concepts

ConceptDescription
ProjectTop-level organizational unit grouping pipelines, ontologies, ML models, digital twins, and storage configurations
PipelineNamed, ordered sequence of processing steps (ingestion → processing → output) executed asynchronously by workers
ScheduleCron-based trigger that enqueues pipelines on a recurring basis
OntologyOWL/Turtle vocabulary defining entity types, properties, and relationships for a project domain
Storage ConfigConnection definition for storage backends with data normalized to CIR format
CIRCommon Internal Representation — normalized record format with source, data, and metadata blocks
ML ModelModel definition linked to an ontology for training and inference by workers
Digital TwinLive in-memory graph initialized from ontology and synchronized from storage
MCPModel Context Protocol — exposes 55 tools for AI agent interaction
Mimir AIP is built in Go for performance and ease of deployment. It runs on Kubernetes and supports a wide range of storage backends, making it an extensible solution for small and medium-sized enterprises.

​
Why Mimir AIP?

Unified Platform

Consolidate data pipelines, ML workflows, and digital twins in a single platform with consistent APIs and tooling

Ontology-Driven

Structure your data and models around formal ontologies, ensuring semantic consistency and enabling advanced reasoning

Cloud-Native

Built for Kubernetes with horizontal scaling, resource isolation, and multi-cluster support for edge and cloud deployments

Agent-Ready

Expose all capabilities as MCP tools, allowing AI agents to configure and operate the platform using natural language

​
Next Steps

1

Quick Start

Get Mimir AIP running locally with Docker Compose in under 5 minutesStart Quick Start →
2

Deploy to Kubernetes

Install Mimir AIP on your Kubernetes cluster with full worker supportView Installation Guide →
3

Connect an AI Agent

Integrate Mimir AIP with Claude Code or other MCP clientsConfigure MCP Integration →

Build docs developers (and LLMs) love

Get started for freeTalk to us