Hard truths, built the structured way..
Structured facts from unstructured documents.
Leonata helps you turn long, messy documents into structured, machine-readable knowledge — without guesswork, hallucinations, or surprises.
It’s not trying to be everything. Just something useful.
If you're working with secure contracts, policies, private clinical notes, or compliance files and need exact answers with clear provenance, Leonata gives you a calm, grounded way to get there.
Built for when we needed clarity.
It’s a command-line script that provides uncontaminated, efficient knowledge graphing and search - without cloud dependencies, costly API calls, or hallucinated data. What’s more, we designed it to run quickly on standard desktop and notebook computers. You do not need GPU’s or expensive large-scale computing resources. Desktop for Mac and Windows.
What Leonata does
Extracts subject–predicate–object facts from unstructured text
Links every output to a source in the original document
Uses rule-based logic and parsing instead of statistical models
Builds a searchable, explainable knowledge graph
Runs locally, works offline, no special hardware needed
Case study
Ray works in the Defence Industry, every week there are Requests for tender pushed out from the Federal Government, they call out specific legislation and often ask the same question in different ways and require the same information across various sections. They require accuracy and their success drives the income of the organisation.
Ray - Head of Software Development, Defence Industry, Australia and United States
“Every single tender document is written by a different person with different language, plus within the one document there is information from the finance, engineering and IT depts, each with their own way of describing things. Nightmare.
Leonata let me accurately map between the documents and cut through the same concept being described in a multitude of different ways.”
Who is it for?
In-house Counsel
You are working on a project that requires analysis of internal classified documents that contain highly sensitive personal information. You need to target a specific area of case law and transpose it against your companies internal policies and procedures plus review previous company responses and learnings, Leonata can be offline and secure, with a fixed monthly cost, it seems to be the perfect fit
AI Pipeline Architect
You have 5-10 pieces of kit strung together to feed data into this app for your client. You’re a contractor and you pay for the tools that save you time out of your own personal pocket. You don’t want to add another, but you see that Leonata will streamline your tasks and reduce your daily workload. Plus eliminate two other monthly costs, and not add compute costs, win win.
Software Engineer
So you’ve heard of Retrieval-Augmented Generation (RAG) and Graph-RAG but you are still starting with contaminated data. Your first point of call is still the LLM and you can’t be sure what the source was and who trained the model. Your client needs an app developed for their customers health data and they need it yesterday, Leonata.
Developer
You spend 3-5 days liasing with your client over each new piece of data you’re running through their new app. It’s come through three LLM’s and you don’t have a medical background so you’re on the phone with them and emailing constantly. They start to question whether ongoing token costs will be sustainable in their business model and launch of the app stalls…you remember Leonata. You’re able to use the terminal to add the critical step and save yourself time and the client money.
Data Analyst
You want to use an LLM to collate the internal risk and compliance records for the mining company you work for but there is a ban on ChatGPT and the internal LLM is still months away, plus you’ve heard it’s just ChatGPT in a skin…you know that Leonata is entirely offline and will comply with your IT dept security policy, plus the finance department will love the fixed monthly cost, no token blow outs.
Health Researcher
You have over 5 million data points for over 200,00 patients which visit your local hospital. The administrator has tasked you with locating which conditions receive the most safety issues for the hospital, how do you complete this task in three months with almost no budget? Leonata of course.
Introducing Leonata
Leonata bridges deterministic extraction and AI: Extract ground truth, then supercharge your LLM workflows.
We have designed a pure, closed system that builds a semantic and syntactic understanding of your text without external influence. Instead of relying on a general-purpose AI that frequently bring in false data, Leonata creates a model that:
Indexes your data based on meaning and structure, not just keyword matches
Preserves contextual relationships between terms without bias from unrelated sources
Generates knowledge graphs that actually reflect how your specific dataset is structured
This means your results are always relevant, reliable, and free from data contamination.
Try It Early. Help Us Shape It.
Leonata is still early. It works, but it’s not perfect. We want to make it more useful, more flexible, and more intuitive—with your help.
We’re actively building:
A web playground with live extraction and graphs
More domain-specific rule packs
A VS Code extension for rule debugging
Guides and docs that don’t assume you have a PhD in semantics
Our Promise
We won’t overpromise or oversell
We’ll be clear about what’s real and what’s roadmap
We’ll respond when you reach out
We’ll treat you as a collaborator, not just a user
We're not here to replace anything. We’re here to offer one more useful tool in your stack — one focused on clarity, trust, and structure.
Get Involved
Request a demo
Whether you're testing it, building with it, or just curious, we’d love to hear from you.
Leonata
Structured answers. Deterministic by design. Built with care.
Help us make it better.
No Hallucinations. No Surprises. Just Reliable Results.
No hallucination
Discrete system. Ensures 100% accuracy by grounding every answer in the provided dataset.
Eliminates risks associated with LLM misinterpretations or creative extrapolation.
On-Demand Knowledge Graphs
Builds a new, dedicated knowledge graph for every new query.
Maximizes precision and relevance by tailoring insights to the exact context of each question.
Security and Privacy
Operates entirely offline—ideal for air-gapped or high-security environments.
Only processes user-provided data, ensuring zero external interference or risk of data leaks.
Efficiency and Scalability
Requires minimal compute resources, making it cost-effective and deployable on edge devices.
Suitable for industries with constrained infrastructure or where high compute costs are prohibitive.
Custom Thesaurus Creation
Dynamically builds and relies on a custom thesaurus based solely on the user's dataset.
Avoids dependency on pre-trained models, ensuring domain-specific and uncontaminated contextually accurate results.
No Training Needed
Skips the expensive, time-consuming process of model training.
Ready to deploy now and deliver insights immediately upon integration.
What’s in the box Leonata?
What’s Under. The Hood.
We’re not inventing the new AI religion, we come from backgrounds in Physics, Plant Molecular Genetics and Golf, we accidently built some symbolic AI in 2019 and thought we’d keep at it. In 2022 Chatty Cathy jumped on the scene and suddenly people had a rag problem and we had a lonely little CLi with no home.
We stand on the shoulders of solid libraries and open standards. We want you to know what’s under the hood. It’s pure and simple. We don’t smoke:
Deterministic Output – Every result is grounded in your dataset, eliminating AI guesswork.
Deep Context Understanding – Extracts meaning and relationships beyond simple keyword matching.
Knowledge Graph Generation – Maps concepts, themes, and structures for fast retrieval.
Be the first to experience the power of deterministic AI
Talk to Us
We’d honestly love your feedback—good, bad, or “this is a useless toy, you’re living a fantasy.”
Email us directly
Request a demo (we’ll just screen-share and show you what it does)
If you build with it, we’ll support you.
If you break it, we’ll fix it (or help you fix it).
If you want to fork it, go for it.
Leonata
No hype. No hallucinations. Just structured data from real documents.
We’d love your help making it better. Pretty please.
Supercharge your LLM
Extracts subject–predicate–object facts from unstructured text
Links every output to a source in the original document
Uses rule-based logic and parsing instead of statistical models
Builds a searchable, explainable knowledge graph
Runs locally, works offline, no special hardware needed
That’s the core. No smoke and mirrors. No promises we can't back up.
-
Developers and software engineers spends hours, days and weeks verifying and validating data that has passed through an LLM. Contacting the client as the technical, local knowledge expert repeatedly can become laborious and onerous to explain the nuances to a client. Natural Language Processing is a science that we have built into Leonata via a semantic model so you can build knowledge graphs from huge datasets in seconds.
-
It’s light. Depending on what you feed it, Leonata can run on a phone or a basic laptop. Add a GPU if you need but this is light, efficient AI so you’ll save money on those NVIDIA chips..
-
If the nature of your project requires you to be air-gapped that is possible with Leonata, our default package calls out for license validation purposes, but we can give you access to an entirely offline product. Zero cloud hosting and zero risk. You store your data where it is secure and you control access.
-
All AI today is trained. Training means a select group of humans once upon a time sat in a room and determined the meaning of words and concepts or a web crawler scraped wikipedia articles to build a thesaurus and a subsequent model. You have zero insight into what built that model and who trained it. The LLM can deteriorate in front of your very eyes. Make your own knowledge graph that is specific to your industry and your know how, using language in the way you intended. Use Leonata’s semantic model to build a Knowledge Graph just for you.
Get to
know us
At Leonata, we’re dedicated to unlocking the creative potential of raw code, transforming it into powerful solutions that elevate your tech journey. Join our dynamic community, where collaboration fuels innovation, and every project pushes the boundaries of what’s possible..
Introducing DRGN: Deterministic Reasoning with Graphs, No LLM Required
Retrieval-Augmented Generation (RAG) made a splash. So did CAG. But what if we don’t want to augment? What if we want to reason?
Welcome to DRGN — Deterministic Reasoning Graph Network — a system for symbolic, ontological reasoning over text, built without transformers, without embeddings, and without hallucinations.
This isn’t RAG’s smarter cousin. It’s a different species altogether.
What Problem Are We Solving?
Modern LLM pipelines are expensive, probabilistic, and fundamentally unpredictable. They're great at generating language, not so great at ensuring factuality, determinism, or explainability.
The typical flow looks like this:
Stuff everything into a vector DB
Hope the embeddings make semantic sense
Pray your prompt pulls the right chunk
Hope the model doesn’t hallucinate the answer
We’ve all been there. Sometimes it works. Sometimes it’s devastatingly wrong.
What if we didn’t have to guess?
Enter DRGN
DRGN builds a knowledge graph at query time using syntactic and semantic cues — not stochastic tokens or black-box embeddings.
It’s built on three assumptions:
Meaning is contextual but structured
Inference should be explainable
No answer is better than a made-up one
Instead of retrieving a chunk and hoping an LLM improvises the answer, DRGN builds a deterministic graph tailored to your query, grounded only in your local data.
How It Works
Text Ingestion
DRGN reads raw text (PDFs, Word, CSVs, Markdown) and performs linguistic preprocessing: tokenization, POS tagging, syntactic parsing.Concept Mapping
It identifies semantically relevant entities and builds a query-specific subgraph using strict ontological rules.Graph Construction
It assembles nodes and edges in a knowledge graph — dynamically, for each query — prioritizing logical consistency over statistical likelihood.Reasoning & Response
The answer is a result of traversing that graph. No transformer. No embeddings. Just paths and logic.
DRGN vs. RAG (and CAG)
TABLE
Who’s Using DRGN (Early Testers)
Defence sector engineers processing classified tenders
Legal researchers tracing precedence across internal memos
Healthcare analysts working under strict data sovereignty policies
Dev teams who are tired of hallucinated output and unpredictable APIs
A Word on Implementation
DRGN isn’t a cloud product. It’s:
A command-line tool
Local-first
Built in Python, runs on Mac, Windows, and Linux
Requires no GPU
We're testing beta builds now. If you’re a dev working with knowledge-intensive text, and you’ve ever yelled “Why is this LLM making that up?”, we want you.
What We’re Still Working On
Graph visualization tooling
Optimized processing for larger corpora
User-defined ontologies and inference rules
IDE integrations and GUI layers
Why We’re Posting This
We’re a tiny team. We’re tired of watching developers struggle with tools that weren’t built for their timelines, constraints, or data policies. If you're building tools for regulated domains, research environments, or just want symbolic AI that isn't dumb, we want your feedback.
You can:
Test DRGN locally
Tell us why it’s broken
Help us shape v1.0
Where to Find It
GitHub is coming. In the meantime:
DM for Mac/Windows builds
Or sign up at leonata.io
Or email us directly: @@@@@
TL;DR
DRGN is symbolic AI for people who want answers they can trust — not improv.
It builds real-time knowledge graphs from your data, offline, no LLM required.
If you're into clean inference and logic over guesswork, let’s talk…
Contact Us
Interested in a project together? Fill out some info and we will be in touch shortly. We can’t wait to hear from you!