GraphNews

RushDB got an update recently that I'm genuinely excited about. It turns scattered data into a graph effortlessly right from the start - autonomously 🤖 tldr → see how it beautifully worked out for Star Wars dataset in comment section below --- Here's the usual situation: - You have `orders`. - You have `users`. Maybe they came from MongoDB collections. Maybe from relational tables with foreign keys. Maybe from CSVs, API responses, or some internal tool that no one wants to clean up properly. The relationships are technically there, but they’re not really "in" the data yet. They live in conventions or in queries. → `order.user_id` points to `user.id` → `customer_email` matches `email` → `companyId` probably connects to `company.id` But until someone wires it up - joins, edges, mapping logic - everything stays loosely connected. RushDB changes how this works. When you import data, it builds a live schema around it. Not a static schema file or a diagram. More like a graph-aware layer that forms as data comes in. Properties become nodes themselves (meta-properties), linked to records by name and type. So when different records share the same kind of property, that structure becomes visible right away. From there, RushDB can suggest relationships. For example: `orders.userId` → `users.id` `payments.orderId` → `orders.id` `tickets.customerEmail` → `users.email` You review and approve these patterns once. After that, new data that follows the same pattern gets connected automatically. That's the part I find most useful. *Relationships aren't something you reconstruct every time you query.* They're created as the data is written. So instead of a bunch of separate collections or tables, you end up with connected records, properties, and relationships that you can explore directly. In practice, it means you can take messy inputs and gradually turn them into a structured, connected dataset without a big upfront modeling step. → Import data. → See suggested connections. → Approve what makes sense. And from then on, new data fits into that structure automatically. Already LIVE for free. Own Neo4j or Aura instance can be brought too. → https://rushdb.com #databases #graphdatabase #dataengineering #backenddevelopment #nosql #mongodb #datascience #softwareengineering #devtools #bigdata #dataarchitecture #techinnovation

Open Knowledge Graphs (OKG) just crossed 2,000 entries -- here are some cool recent additions: 🖼️ Lightweight Image Ontology (LIO) — a small OWL ontology for relating image descriptions to visual parts. Useful for anyone doing image annotation or computer vision provenance. Margaret Warren and Pat Hayes https://lnkd.in/egHH-RgV ♟️ Universal Open Game Theory Ontology — modular OWL + SHACL for game-theoretic semantics. Brand new (repo created last week). Worth watching. https://lnkd.in/earFmqcR 🔒 NORIA-O + DSecO — a pair of ontologies from Orange. NORIA-O models IT network operations and events; DSecO models DNS data for reasoning over it. Useful combination for network security semantics. https://lnkd.in/eG6_T4ud and https://lnkd.in/ez3DvG-c ⚖️ Akoma Ntoso for EU (AKN4EU) — the structured XML/vocabulary format for EU legislative documents, maintained by the Publications Office of the EU. Foundational for anyone working on legal linked data in Europe. https://lnkd.in/e7DknN33 📢 IAB Content Taxonomy — the industry-standard classification for digital content used across adtech for contextual targeting and brand safety. Now in the catalog. https://lnkd.in/evKMWZj4 ♻️ dpp: namespace — an open JSON-LD vocabulary for the EU Digital Product Passport. https://lnkd.in/exXdjQvQ 🍽️ LanguaL — multilingual food description thesaurus used to describe and retrieve information about food. A classic that wasn't in the catalog until now. https://lnkd.in/e6whdhJT 🏛️ FISH Evidence & Monument Types Thesauri — UK heritage vocabularies covering physical remains, buried monuments, buildings, and structures. Solid controlled vocabularies for the archaeology/cultural heritage crowd. https://lnkd.in/eCCd5GdJ and https://lnkd.in/eRbEHURw 🔗 Cointegrity Web3 Glossary — 2,000+ terms covering blockchain integration and digital asset regulation. https://lnkd.in/eQd7XjuU The catalog now covers 2,011 ontologies, vocabularies, and graphs. openknowledgegraphs.com

SHACL constraint validation + OWL 2 DL reasoning + visual graph editing — all running in your browser, zero install. Ontosphere v1.

Vercel's Eve framework makes durable agents easy - but agents still forget. Here's how I gave one persistent, graph-native memory using the Neo4j Agent Memory Service (NAMS), built around a real National Parks trip-planner called TrailGraph.

Part 1B of the series. Where the first part (1A) made the architectural argument, this post shows how we wired the first iteration of it…

This new series are following Meta Knowledge Graph post earlier. Part 1A of a series. An opinionated view of how a self-learning…

How Holonic Architecture Escapes the Ontology Unification Trap

Property graphs have no reasoning engine. Most people don't notice because they don't know what it is. A reasoning engine is a system that derives facts you never explicitly stored. For instance, the sub-classing idea is transitive, it knows that if Mammal is a subclass of Animal, and Dog is a subclass of Mammal, then Dog is also a subclass of Animal. This is revealed without you asserting it. Given a rule that says "a person who authored a paper at institution X was affiliated with X," it infers the affiliation. You assert the minimum; the engine closes the world (infers the unspecified). This is what OWL entailment and RDFS semantics gave RDF triple stores in 2004. Formal, auditable, reproducible inference. Load a consistent ontology, run a reasoner like HermiT or Pellet, and your graph knows more than you told it. Property graphs (Neo4j, Memgraph, Kuzu, TigerGraph...) have none of this. You write Cypher. You get back exactly what you put in. The graph cannot derive a single fact that wasn't explicitly asserted. The schema is flexible and the tooling is excellent, but the semantics stop at the storage boundary. If necessary people used to create it in the business logic (layer). Nowadays everyone is wiring LLMs on top of property graphs and calling it "AI-powered knowledge reasoning." Well, it isn't. The LLM is guessing the inference layer inconsistently, expensively, without auditability, and without any guarantee of logical consistency. It depends on the (size of) model, how it fits the data, the language and more. A reasoning engine gives you closed-world guarantees. An LLM gives you a confident guess. The reasoning gap in labeled graphs is not a footnote. It is the central unsolved problem in production knowledge graph deployments and almost nobody in the LPG ecosystem is talking about it. It's inconvenient. Don't get me wrong, I worked with and for Neo4j, with and for TigerGraph. I like LPGs and advise customers to go for it, if I am convinced it's the right choice. What I try to emphasize is what you get and what you think you get out of LPGs via LLMs. OWL Implementations: https://lnkd.in/eVXHkjYn #Semantics #KnowledgeGraphs #Reasoning | 16 comments on LinkedIn

Databricks built a new context layer for AI agents and named its engine OntoRank. The first half of that name is the claim. The second half is the mechanism. Databricks drew the comparison itself, presenting OntoRank as PageRank for your company's data. The offer is the one every major platform is making this year: infer your meaning, rank it, hold it. Databricks just named its algorithm. 🕸️ The graph Google already built Google's PageRank is 1998. It ordered web pages by how many other pages linked to them, popularity standing in for authority. In 2012 Google built something else. The Knowledge Graph. Things, not strings. A structured layer of typed entities and relationships, on Freebase, acquired two years earlier. Popularity can rank a page. It cannot tell you what a thing is, or where one concept ends and the next begins. By 2016, a knowledge panel or direct answer appeared on roughly a third of Google's 100 billion monthly searches, from the graph, not ranked links. 🎯 Google's path, run in reverse Google moved from ranking to structure: it began with popularity, then decided, thirteen years ago, that popularity was not meaning. OntoRank runs that in reverse. It takes the 1998 idea Google judged insufficient for meaning, ranks your data by authority and usage, and calls it an ontology. It weighs more than frequency: who authored a definition, how widely it is used, how recent. But each is a signal it ranks, not a rule it enforces. It infers which definition looks authoritative; it does not decide which is right. Deciding what a concept includes, excludes, and why is the authoring step, and a ranking cannot perform it. A popularity score is not an ontology. Naming it OntoRank does not make it one. Popularity is not provenance. 🔑 The part the vendors left behind The deeper lesson is not about ranking. Google did not rent its Knowledge Graph. It acquired Metaweb and its Freebase knowledge base, built a governed, structured asset on top, and has compounded it for over a decade. I have watched an owned meaning layer outlast the vendor whose tools ran it. I opened this series on exactly this: one of Google's AI advantages is the graph it owns, not the model it rents. Take the inspiration, but the right part: the posture, not the scale. You do not need Google's graph. You need the few definitions that carry your business, authored and owned, with automation on top. Model what matters first. The platforms are selling the reverse: a meaning layer they infer, rank, and hold for you. You get the snapshot. They keep the asset. 💡 The verdict The vendors took half of Google's playbook. They built a real knowledge graph, ranked it on a 1998 idea, then named the part they did not build: an ontology. They left what made the original graph work: authored structure, and ownership that compounds. Copy the half that compounds, not the half they sell. #AgenticAI #AIGovernance #KnowledgeGraphs #Ontology #TrustworthyAI

On data sovereignty, knowledge infrastructure and meaning. Three emerging trends so far, in 2026. Data sovereignty, knowledge infrastructure and meaning. Five products promise to deliver. But do they?

A semantic layer is a system of technologies and techniques that organizes and maintains a consistent and unified representation of data from various sources that is interpretable by humans and machines. MIT CISR research indicates that leaders are feeling pressure to increase their investments in semantic technologies and techniques that can store metadata and instantaneously make data assets more useful to business users, AI models, and AI agents. This briefing describes a target state for a semantic layer and offers three next steps that leaders can take to ensure their investments in semantic technologies and techniques pay off.

Getting classical music events out of orchestra websites and into a knowledge graph using ontology as a lens.

Using Google graph technologies, Yahoo's Seller Agent platform is a model for other industries striving to build a trusted agentic system of action.

🚨 Yarchi just published an awesome guide on building an AI "second brain" using Claude and Obsidian Based on @karpathy's LLM Wiki concept, the idea is to stop starting from a blank chat window and start building a compounding knowledge graph. Obsidian handles the local storage. Claude acts as the brain. I wrote down the condensed notes on how to wire them together: The Stack: → Install Obsidian (your free, local text storage) and Claude Desktop Pro (the engine that reads and writes files). The Bridge: → Install the "Local REST API" plugin in Obsidian, copy the key, and use Claude’s MCP framework to connect them (claude mcp add-json obsidian-vault...). The Context: → Ask Claude to interview you. Have it ask about your goals, projects, and working style, then save it all to a CLAUDE.md file so it never forgets who you are. The Structure: → Create specific project folders (Inputs, Process, Outputs, Feedback) and open them as isolated vaults so Claude stays focused on one job at a time. The Automation: → Save repeated workflows as markdown "skills" and use Claude's schedule tab to trigger daily vault maintenance. Voila! You end up with an AI that organizes your files, summarizes your notes, and remembers your entire context. Full guide from Yarchi here → https://lnkd.in/e6yyWk6d | 40 comments on LinkedIn

Andrea Volpini says the ontology is becoming the control plane for AI agents. Sergey Vasiliev warns we’re scaling unvalidated meaning into…

I keep getting the same question from CIOs: "How do I justify investing in a context layer?" After building alongside dozens of enterprises this year, five metrics keep showing up: 𝟭. 𝗧𝗶𝗺𝗲 𝘁𝗼 𝘃𝗮𝗹𝘂𝗲. Building an agent takes five minutes. Giving it the business context to actually work? That's where teams get stuck. One customer spent 12 months in testing hell because their CEO-facing agent kept generating incorrect results. A context layer helps you bootstrap that foundation and compress shipping time from months to days. And as more people across the org start building on top of it, the time to value drops even further. 𝟮. 𝗔𝗰𝗰𝘂𝗿𝗮𝗰𝘆. In personal workflows, close enough is fine. In enterprise workflows, if your agent doesn't hit 70-80% accuracy, the business will stop using it. The context layer lets you simulate before you ship. Take the 100 questions your agent is most likely to get. Run them against your context. Find the gaps, fix them, then go live. We're seeing accuracy jump 2-10x on agents that go through this process. 𝟯. 𝗖𝗼𝘀𝘁. Without context, agents burn through tokens because they traverse everywhere and don't know where to look. Context gives them a map. One customer cut token costs by 75% just by giving their agents context. The agent was able to stop guessing. 𝟰. 𝗔𝘂𝘁𝗼𝗻𝗼𝗺𝘆. One retail customer shipped an agent that could handle 60% of decisions on its own. Every time a human stepped in on the other 40%, that correction became context the agent could learn from. Each override makes the next decision better. That's how you push 60% toward 90%. 𝟱. 𝗣𝗼𝗿𝘁𝗮𝗯𝗶𝗹𝗶𝘁𝘆. Every CIO I talk to raises this one. Nobody wants their enterprise context locked into a single agentic framework — that lock-in is itself a form of technical debt. We built something called a context repo, basically a GitHub repo for enterprise context. You own it, you version it, and you can move it if you need to. The context layer is becoming table stakes for enterprise AI. What other benefits are you seeing?

What it takes to build the learning loop Satya described, and why most organizations leak it before it compounds. Satya Nadella’s article on what happens when context compounds went viral last week on X, and it’s easy to see why.

A revised and more synthetic version of the original article, incorporating feedback from readers who noted that the destination only became visible near the end. Preamble Following discussions with several readers, I realized that my previous article attempted to cover too much terrain at once.

RDF 1.2, Reification, and the Return of Ontology to Language

Most databases store records. FlureeDB stores knowledge — and the proof of where it came from. A walk through the features that make a knowledge graph verifiable: immutability, provenance, per-triple security, time travel, and more.

By the end of this blog post, you are going to experience two epiphanies about databases. However, you are going to have to do a little more work than just read. You are going to have to stop and a…

While industry veterans keep debating RDF vs. LPG, both camps seem to agree on one important thing we need in the agentic era: #Ontologies They have been around for decades. But only now does the idea feel like it is compounding and becoming practical at scale - especially after Databricks recent ontology-focused announcement. The old debate usually sounds like: RDF gives you explicit semantics, predicates, ontologies, and reasoning. LPG gives you a practical developer experience, flexible properties, graph traversal, and a model that feels natural for applications. Both are valuable. But for agents, the real question is not “which graph model wins?” The real question is: How does data become self-describing enough for an agent to safely query it? At RushDB, I blended the best parts of both worlds through LMPG - Labeled Meta-Property Graphs. The idea is simple: when data is written, RushDB does not only store it. It also creates a dynamic metadata layer from it. RushDB creates meta-properties on write from any JSON or CSV, inferring property names, types, labels, and relationships. These meta-properties connect records by the pair of name + type. Scalar values stay inside Records, preserving the familiar LPG-style developer experience. Meta-properties become a dynamic schema layer that evolves every time the data changes. This gives you something that sits between RDF and LPG: * explicit property identity * dynamic schema discovery * familiar record-like storage * graph-native traversal * ontology-aware structure without forcing users to manually design an ontology upfront The important part is what this means for agents. Agents should not have to guess what fields exist. They should not hallucinate schema. They should not generate queries against imaginary properties. They should be able to ask the database: “What labels exist?” “What properties exist?” “What values are actually present?” “What relationships are available?” “What can I safely query next?” That is the direction I care about: databases that do not only store data, but also explain the shape, limits, and possibilities of that data. For agents, the database should not just be a persistence layer. It should be a grounding layer.

A field guide to what it is, what it is not, and where it fits in your AI architecture.

Jakki Geiger, CMO at Arango in conversation with Ravi Marwaha, Chief Product & Technology Officer at Arango

If you want to become good at building AI memory systems, Then study these 5 knowledge graph case studies: 1. Your Agent's Reasoning Is Fine. Its Memory Isn't.    → https://lnkd.in/et8fmmBQ     2. Building Agentic GraphRAG Systems    → https://lnkd.in/eJG5JFm4     3. Inside Neo4j's Agent Memory    → https://lnkd.in/eyr94A-P     4. Stop Chasing the Perfect Ontology    → https://lnkd.in/eE-PZbPF     5. How to Keep Your AI Agent's Knowledge Graph Clean    → https://lnkd.in/dVkT7vzB What else should make the list? P.S. Want more deep dives like this? Join 40,000+ engineers reading Decoding AI Magazine. Link: https://lnkd.in/dgKZFc5j | 22 comments on LinkedIn