— Sovereign AI · Open-Source Infrastructure

The Open-Source
Analytical Stack

Geoeconomic intelligence requires verifiable foundations. Merlin Intelligence open-sources its core reasoning systems under the MIT licence—enabling practitioners, institutions, and autonomous agents to detect knowledge singularities, map cognitive structures, and construct formal domain ontologies without dependency on closed AI models.

How the Stack Operates · Triple Engine Workflow
Step 01
Heterogeneous Corpora
Primary treaties, regulatory frameworks, central-bank statements, and trade flow data aggregated into chunked corpora.
Step 02 · eigenmind
Singularity Extraction
Computes eigenvector centrality and spectral graph decomposition to identify high-density conceptual nodes and expose structural pivots before market consensus.
Step 03 · olaf (MCP)
Formal Ontology Construction
Orchestrates LLM agents via 22 MCP tools to assert verifiable OWL/RDFS triples backed by Oxigraph and Qdrant.
Step 04 · CogMaps
Unified Visual Interface
Brings eigenmind and olaf together in one workspace over the Qdrant-stored corpus, blending classic RAG with graph-based metrics and ontological overlay.
eigenmind MIT Licence Python Graph Spectral Analysis

Singularities. Structural Pivots. Sovereign AI.

Autonomous Knowledge Singularity Detection Engine

Standard generative AI models excel at summarising widely distributed explicit knowledge, but remain blind to tacit trade-offs, inflection points, and structural contradictions. eigenmind identifies singularities through graph computation—isolating the structurally unusual, high-density nodes that standard search overlooks by computing eigenvector centrality and spectral density across the corpus graph.

Core Capabilities

  • Singularity Mapping: Computes topological density to reveal latent regime changes in geopolitics, trade, and financial corridors.
  • Eigen-Decomposition Core: Derives eigenvector centrality and Laplacian spectral embeddings to surface nodes that are structurally load-bearing across multiple analytical framings.
  • Causal Edge Weighting: Evaluates edge significance by semantic proximity and verifiable causal taxonomy rather than superficial word co-occurrence.
  • Reproducible & Transferable: Complete source code delivered under MIT license with no external cloud lock-in.

Technical Architecture

  • Algorithmic Base: Graph spectral analysis (eigenvector centrality, Laplacian embeddings), semantic embedding projections, and causal dependency graphs.
  • Primary Output: Ranked singularity scores, structural pivot maps, and early warning radar metrics consumed by CogMaps for hybrid ranking.
  • Target Environments: Sovereign intelligence units, risk analysis desks, and quantitative strategy research.
View eigenmind on GitHub → github.com/merlin-intelligence/eigenmind
olaf MIT Licence MCP Server OWL / RDFS Oxigraph + Qdrant

Ontology Learning Agentic Framework

Model Context Protocol (MCP) Server for Collaborative Knowledge Engineering

olaf bridges generative LLM agents with formal knowledge representation. Rather than generating fragile one-shot graphs, olaf exposes a standardized Model Context Protocol (MCP) server that empowers agents to read text chunks incrementally, identify candidate concepts, query existing graphs, resolve semantic duplicates, and assert formal OWL/RDFS triples with persistent validation.

Dual-Storage Architecture

  • Oxigraph RDF Triplestore: Native HTTP triplestore managing formal OWL/RDFS statements, named graphs, SPARQL queries, and transitive reasoning.
  • Qdrant Vector Engine: Manages document chunk ingestion with status tracking alongside dedicated olaf_concepts_{id} collections for real-time concept semantic search.
  • FastEmbed Embedding Layer: Automated multi-lingual embedding pipelines for high-precision concept matching.

Agent Integration & Tools

  • 22 Native MCP Tools: Complete tool suite for chunk inspection, concept creation, synonym linking, relation assertion, and hierarchy validation.
  • Stateful & Resumable: Agents can pause, refine, or resume ontology construction across distinct sessions without context loss.
  • Client Compatibility: Direct plug-and-play support for Claude Desktop, Claude Code, Cursor, and any MCP-compliant agent host.
View olaf on GitHub → github.com/merlin-intelligence/olaf
CogMaps MIT Licence Qdrant RAG + Graph + Ontology

One Interface. Two Engines.

Visual Interface Unifying eigenmind and olaf

CogMaps is the visual interface that brings eigenmind and olaf together into a single analytical workspace. Documents are stored and indexed in Qdrant, and every query combines classic retrieval-augmented generation with eigenmind's graph-based centrality metrics and olaf's formal ontology as a live overlay—so results are ranked by relevance, structural significance, and verified conceptual relations at once.

Core Capabilities

  • Unified Workspace: A single interface over eigenmind's singularity graph and olaf's formal ontology—no switching between tools to move from a document to its structural context.
  • Hybrid Retrieval: Blends classic RAG over Qdrant-stored document chunks with graph-based ranking driven by eigenmind's eigenvector centrality and singularity scores.
  • Ontological Overlay: Superimposes olaf's OWL/RDFS concept hierarchies directly onto retrieved documents and graph views, grounding every answer in verified formal relations.
  • Reproducible & Transferable: Complete source code delivered under MIT license with no external cloud lock-in.

Technical Architecture

  • Document Store: Qdrant vector store shared with olaf for chunk ingestion, embeddings, and semantic search.
  • Retrieval Engine: Hybrid pipeline combining semantic similarity search with graph-based metrics pulled from eigenmind.
  • Ontology Integration: Queries olaf's Oxigraph triplestore to overlay formal concept relations on retrieved results.
  • Primary Output: Ranked, ontology-annotated answers alongside interactive graph and concept visualizations.
View CogMaps on GitHub → github.com/merlin-intelligence/cog-maps
Our Methodology

Why Open-Source Infrastructure?

Geoeconomic decisions in energy corridors, capital reforms, and strategic supply chains cannot rely on black-box heuristics. Our open-source commitment guarantees three pillars:

1. Sovereign Reproducibility

No external API dependencies or proprietary vendor lock-in. Code, ontologies, and reasoning graphs remain entirely within the custody of the institution.

2. Verifiable Evidence Chains

Every asserted relation and discovered singularity links back to primary legislative, financial, or spatial source data through formal RDF triples.

3. Human-Machine Complementarity

Large models provide broad lexical extraction; human experts and formal ontology protocols validate decisive strategic insights.