I just released the book. If you’re not following me, actually, I have a decent amount of books about different topics, but all of them are mainly about the agentic memory and the data structure around the memory. And I was focused on the privacy-preserving artificial intelligence. So the majority of the data and processing happens on your phone or on the mesh of your private devices, and maybe some LLMs even hosted on your phone. And it’s a separate discussion for the small language models and how they will transform the privacy and why it matters for the individuals.

https://leanpub.com/ladybugdb

Sovereign Agentic AI (Volodymyrs View) is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

And I talk about the sovereignty in my previous chapter of the podcasts. But what’s special about this book? First of all, it’s describing how to work with the property graphs with the OpenCypher. And Cypher is the query language for the property graphs that was invented by Neo4j, and right now we have it in a Ladybug with interesting features like the relational and node tables with the strict types and all other things. So the Ladybug is an amazing database that’s unlocking the RAG and memory on the user devices for the private agents.

And it came also with seven simple ideas that I was keen to emphasize in the book. The first one is the meaning of the private AI. And it’s the main topic what we’re talking in this podcast—how to have private, sovereign AI that you could rely, you could own, and you could control. And make sure that your data is not contributing to somebody else’s training dataset that will discriminate the group of people to whom you belong. And taking into the account all the madness that happening right now in US War Department, or how they call it right now, and the big LLM providers, we could see how the things could horribly go wrong in this space.

Another interesting idea that I try to bring is that we need something more than memory or we need something more than RAG, first of all. So because RAG is the retrieval-focused process and it’s much more than the memory, right? It’s much less than the memory, let’s say. And yeah, so then I’m describing how to take the property graph model and explain the basics of the graphs and then I try to lead them to the heavy lifting topic of the AI-empowered data structures like the hypergraphs, the metagraphs and why they matter for the agentic workforce and why we still could use the technology that was not designed for it but still able to do the job. After that, we switch to the topic of implementing the semantic spacetime. It’s the special kind of ontology that allow us to describe the complex object around us and offer us something much powerful than simple graph embeddings and similarity and bring the similarity as the graph structure.

And in the same time, I would say that we construct the metagraphs based on the semantic spacetime. That are quite powerful concept of nesting the information and creating the complex relations. And actually, I have the working strategy of the big graphs that make the hypergraph and metagraphs working with the simple Cypher and the property model. Then we switch to the deeper topic: we model entire memory for the agents that I describe in my separate book. So I sacrifice my entire book to make this book and just take the distilled summary of it and bring it to your table as the working database schema and code, line-by-line of the schema, and explaining how it works, how to use it, what kind of entities we have, what kind of properties of these entities we have. So it’s really special book, I would say.

And then we will go deeper to some technology advantage. First of all, the Ladybug have the side project, the Ladybug Memory, that have its own memory model that it’s worth of give it a try. It’s built on top of the Ladybug and have own view on the information and we just try to demystify it. Also I describe some super advanced features that no one of the graph databases have, like the ability to have the extensions, work with JSON, work with graph algorithm, work with the relational databases with the power of the Ladybug, and all other things. So it’s really deep topic from one side of the database that nobody heard yet, but the database that could really heavily change the landscape of the graphs. And from the other side, it’s really the deep topic what the agentic AI needs to move forward.

And for sure I try to explain how the conversational memory and memory of the agents could be coupled with something that I called the promise graphs. I also explain the promise graphs and give the property model and ontology of the promise graph that could be used to track the multi-agent and agent-to-agent communication. So it’s quite unique book with, let’s say, 20 euros price or something like that, and then really go deep on so many topics that go beyond. And for sure I intentionally haven’t mentioned the context graphs because if you have the property graphs and agentic memory, you have the context graphs for free and you have the context management for free in completely different philosophy, actually. So it’s some kind of combination.

And I haven’t talked actually deeper that the promise graphs open us the door to something that I call the social memory, where the agents memorize with whom they could interact on which tasks and who is the good guys, who is the bad guys, and all other things. So it’s still open-ended book that captures so many concepts that for me it was hard to stop. But the most important thing is that if majority of my books was the abstract philosophical explanations of the different kind of graph model and architecture, this book is 85% hands-on. So it’s not just explaining the concept, it’s just go and build it with the property database with the concrete fields. That, for example, this memory contains so many interesting patterns like the temporal axis, the temporality, and the different dimensions of the time, the clustering on the kind and the context, the clustering of the graph based on the layers, introduce the layered graph rules and all other things.

So this book, it’s somehow the quintessence of all the research that I made in the last two and a half or even three years. And right now with the Ladybug, I really made it happen. So we still have the place where we could model it and store in a more type and data-driven way, because the Ladybug, it’s the special kind of database that have the node and relational tables that actually have the fixed structure. So it’s not so weird as our favorite Neo4j where you could actually on-the-fly create the relations and nodes that was accidentally mistyped or just broken.

So give it a try. The Ladybug is amazing database. Give it a try to read the book because it’s really special. Take a look to the memory structure, take a look to the promise graph concept. And I still keen to talk about the cognitive processing and how to digest the bigger piece of information to more structured memory and why it should happens gradually. And why it should happens gradually especially on the devices that have the limited processing capacity while your app is still open. It’s all complex questions that deserve a couple of good chapters in a book, or maybe book itself. Because the structure of the data is not enough. We needs to have the cognitive model and concept model on top that allow us to build the processes of all the things. And the processes sometimes is more important because even the reconstruction of the memory from the persisted storage for the human, for example, it’s completely magical and complex reconstruction process where you damage the information a bit and your reconstruction is really depends from your mood, the level of sugar in your blood, and the stars on the sky. That’s why sometimes it’s too creative and sometimes it’s so unreliable that even this process itself deserve the separate book.

And this book I already try to write and it’s 85% ready because I have the memory model, but I skip all this human-like flaws of the memory and why we needs to invent something more reliable than the human memory but still could mimic some good things from the human memory. And this book is still waiting of my attention, to be honest, and I feel a bit ashamed that I could not finish it for more than six months already. But it have so many contradictive material that I needs to process and make my own opinions, not only from the technical side but from the side of the neurology and cognitive science that it’s really hard to finish, to be honest. But I keeping try hard enough. So give it a try to the Ladybug, read my book, and say what you think about it and if you building the agents, you could build the agents that remember now much better than before. So see you next time. Ciao!”

Would you like me to create a technical glossary for the specific terms mentioned in this transcript?

Get full access to Sovereign Agentic AI (Volodymyrs View) at volodymyrpavlyshyn.substack.com/subscribe