Vasa SerafinChat, on the web and in the app
The same assistant, reachable from a browser at nexusagi.thetooltuckshop.rs or from the free Android app.
Honest, plain-English
What NexusAGI can do today
This page exists so you never have to guess. It says what NexusAGI is, what it actually does right now, what it does not do yet, and how the free access works — no marketing language, no invented statistics.
What NexusAGI is
NexusAGI is a rule-based, continuously learning assistant that runs on the operator's own hardware. It is not a large language model, and it does not send your conversation to a cloud AI service to generate its replies. It answers from a knowledge graph it builds and edits itself, following rules it also builds and edits itself, on a machine its operator owns. When it does not know something, its rules are designed to say so rather than guess a plausible-sounding answer.
What it does today
Teach it, and it remembers
Tell it a fact and it stores it in its own knowledge graph, to draw on in later conversations.
Learns from feedback
Correcting it or confirming an answer feeds back into how it responds next time.
Sessions and usage tracking
See which devices are signed in, follow a session from your phone, and check your usage.
Definitions and explanations
Across the knowledge branches it holds — general knowledge, science, and more — built from what it has been taught and has ingested.
Maths, science and language curriculum
Curriculum-style knowledge across arithmetic, algebra, geometry, physics, chemistry, biology and English.
Code questions
For the programming languages it currently holds rules for (including C#, Java, Python, JavaScript, PowerShell and Kotlin).
Terminal follow, in the app
Start a session on your computer and follow its terminal from your phone — the app can send messages back but cannot run anything on that computer.
The numbers
Measured directly from the running system, not estimated, and rounded down so nothing here overstates it:
Over 9,400 rules
The logic NexusAGI reasons and answers with — its own rules, not a trained model's weights.
Over 150 rule flows
Named, checked chains of rules — the structured procedures it follows for a given kind of question.
Over 2,000,000 knowledge items
Nodes in its own knowledge graph — facts, definitions and relationships it has been taught or has ingested.
Rule and flow counts include disabled rules that are excluded from active use. Source: the
live system's own /api/health, /api/rules/effect-coverage and
/api/rules/graph/flows endpoints, measured 6 September 2026.
What it does not do yet
- No image generation. NexusAGI does not create pictures. (See "What's coming" below.)
- No web browsing by default. It answers from its own knowledge graph, not by fetching live pages, unless a specific feature is explicitly enabled for that.
- Benchmark scores are being re-measured honestly. An earlier internal check found that some of NexusAGI's own claimed benchmark results did not hold up under a fair re-scoring. Rather than publish a number we are not confident in, we are re-measuring properly — so this page does not quote an accuracy percentage. When there is a number we trust, it will appear here with how it was produced.
How free access works
Website
1 free day of chat after you sign up, no card required.
Android app
7 free days from the first time you open the app.
Then
The subscription is AUD 30/month, billed through Google Play, and unlocks the web chat, the Android app and the CLI together.
Privacy, in one paragraph
NexusAGI runs on the operator's own hardware; your conversation is not relayed to a third-party AI provider to be answered. The Android app declares exactly what it collects and why, including the owner-verification camera/microphone feature it is not yet using in this build. Read the full detail in the NexusAGI privacy policy.
What is coming
These are plans on the development queue, not promises of a date:
- Its own image generator — grounded in what the knowledge graph already knows about an object's parts, shape, colour and size, rather than a third-party image model.
- Arduino and temperature-sensing support — reading sensors and working with microcontrollers as a manifest tool. This is planned to land last, after the rest of the current queue.
- Automatic updates — the CLI checking for and installing new versions itself.