AGI, SLM and NovaCopilot

What is AGI?

AGI (Artificial General Intelligence) means a system that can understand, interpret and solve problems in a stable, cross-disciplinary way — with consistency, logic and source evidence. What AI companies are trying to achieve with their LLMs, NovaCopilot already delivers within law through an SLM (Source-Linked Model): determinism, zero hallucinations, full provenance and identical answers.

• understands, interprets and solves problems stably across disciplines
• builds answers on consistency, logic and source evidence
• within law this is achieved through a source-linked model, not a generative one

What is an SLM?

An SLM (Source-Linked Model) is a deterministic source engine that does not generate free text but retrieves locked, verified source blocks from a controlled register.

• does not hallucinate and does not guess
• uses no external LLM models
• gives identical answers every time
• builds answers on primary sources: statutes, EU articles, case law and regulations

NovaCopilot in practice

NovaCopilot is the SLM installed as a PWA — an evidence-carrying legal source engine that runs offline on phone and desktop, with close to 21,000 verified data points from the platform's own register.

• offline search with no cloud dependency in the lookup itself
• no third-country transfer and no model training on user data
• same answer, same source, every time — traceable under the AI Act

Open NovaCopilot

The difference from a generative model

A generative model weighs the probability of the next word. NovaCopilot retrieves the locked source line. Every answer can therefore be audited against the register — line by line, with fingerprint and read date.

Common questions

Is NovaCopilot a chatbot?

No. NovaCopilot is a deterministic source engine. The answer is assembled from cited data points in the register, not from generated text.

Does this meet the AI Act's requirements?

The engine is built for traceability and transparency: every answer carries its source, its read date and a machine-readable fingerprint, and no answers are freely generated.

What are the answers based on?

The platform's own register: Swedish statutes and rulings, EU acts at article level, US case law, ESG rules and question pages — close to 21,000 data points.

How does an SLM differ from a RAG setup?

In a RAG setup a generative model writes the answer from retrieved text. An SLM hands over the source line itself. Nothing is rewritten, so the answer can be compared word by word with the register.

Can the answers be used in a legal matter?

The answers are cited extracts with read date and fingerprint, meant as source material and traceable references. They do not replace assessment by counsel.

How are the AI Act's transparency requirements handled?

Every data point carries origin, read date and version, the engine is documented in the model card, and it is always stated that the answer comes from a register rather than a generative model.

Is personal data processed when I search?

Search in NovaCopilot runs locally in the app. The query is not sent to any external language model and is not used for training.

How often is the register updated?

The register is updated continuously. Each data point carries its own read date, and freshness is published in machine-readable form so both humans and agents can see how current a line is.

What happens when a source changes?

The line gets a new read date and a new version. The previous version is not silently removed; it can be followed through the evidence chain.