One company, three AI directions: why Kvantim is building an ecosystem
A personal AI, a business supervision layer and a trainable transformer are not the same product. They are three answers to three gaps in how useful AI systems are built.
A person who uses AI every day needs continuity. They do not want to explain the same project and preferences in every new session. They also need to know where that personal context lives and how to correct or remove it.
When people say "AI product", they often mean the model. Pick a model, add a chat interface, connect a few tools and the product is done. Real use quickly shows why that is not enough.
A business needs something else. It may be willing to let AI prepare a report, update a system or draft a customer response, but not with unrestricted access. The business needs permissions, approval gates, credential protection, recovery and a record of what happened.
An organisation with specialist knowledge may have a third need: a language model that can be trained and evaluated around a particular domain instead of treating a general model as a fixed black box.
These are connected problems. They are not the same product. That is why Kvantim is building three complementary directions inside one company.
Kvantia AI: personal context and local control
Kvantia AI asks what a personal AI should look like when the relationship is organised around the user rather than a central conversation service.
Its direction is local-first. The product is being designed to run on hardware the user controls, retain the context the user chooses and make memory visible enough to inspect, correct and forget. Teaching should be deliberate: an example, correction or preference becomes part of the system because the user chooses it, not because every private conversation silently becomes training material.
This direction also accepts a practical trade-off. A smaller local system is not the largest cloud model. Its reason to exist is personal continuity and control, not a claim that it wins every general benchmark.
The current status is a working prototype. Persistent memory, teaching and broader distribution remain separate milestones. We state those statuses because the difference matters.
Kvantia Harness: controlled execution for business
Kvantia Harness solves a different problem: how do you let an AI worker perform real digital work without giving it free rein?
Harness is not the language model. It is the supervision layer around the work. It accepts tasks, exposes only permitted capabilities, pauses sensitive actions for a person and records the complete run.
That distinction is important. A model may be capable of deciding that an email should be sent. Harness is responsible for whether sending is allowed, whether this recipient is permitted, whether a person must approve and what appears in the audit trail.
This is where Kvantim learns about AI in real operations: which controls teams need, where workflows break, what evidence makes a result reviewable and how a system should recover when a source is unavailable.
Harness is in early access. Its exact status is shared across kvantim.com and kvantia.app rather than rewritten for each page.
Kvantim LLM: a transformer that can be trained further
Kvantim LLM is the company's classic language-model direction.
It starts from an open-source transformer model base and is being rebuilt for further training and specialisation. Before release, the important questions are not the name of the model family or the number of planned variants. The important questions are concrete:
- Which open-source model and licence form the starting point?
- What has Kvantim changed?
- Which training method changes the model?
- Which data may legally be used?
- How is the result evaluated?
- What can a customer deploy, move or continue training?
Those questions keep "trainable" from becoming a slogan.
Kvantim LLM is technically distinct from Kvantia AI. Harness is distinct from both because it is the control layer around AI work. One ecosystem does not mean one architecture hidden behind three names.
What the three directions share
The products share a company, team and discipline.
They can compound the same expertise in AI evaluation, failure analysis, privacy, local and controlled deployment, product security and honest status reporting. They can also create different relationships with the market.
Harness brings the company close to business workflows. Kvantia AI develops the personal and local direction. Kvantim LLM builds trainable transformer capability. Learning in one direction can influence how the company evaluates and ships another, without pretending that an integration already exists.
That last boundary matters. A future possibility is not current traction. A working prototype is not a generally available product. An open-source base is not automatically proprietary intellectual property. We want the ecosystem to be stronger because those distinctions are visible.
Why build it as one company?
Building three directions creates focus risk. A small company can spread itself too thin. The answer is not to deny that risk. The answer is milestone discipline.
Each direction needs a named owner, a buyer, a status, an evidence requirement and a clear next gate. A direction expands only after it proves the milestone in front of it.
The advantage of one company is shared learning and more than one route to value. The cost is the need to be unusually clear about priorities. That is why our public product pages and investor material use the same status system.
Kvantim is one company. Kvantia AI, Kvantia Harness and Kvantim LLM are three focused directions. Together, they represent our view that useful AI needs personal context, controlled execution and a real path to adaptation.