A proprietary model that answers from its own texts, names its source for every answer, and says plainly when it does not know.
Open the live modelSphuratta AI is a proprietary artificial intelligence model developed and owned by SPHURATTA PROJECT CAPITAL INC. At its centre is an original core architecture, designed and built in-house; the design of the core is a trade secret and is not disclosed. Around the core, the model's English language model is trained entirely from scratch on the company's own licensed corpora — no third-party models, weights, or APIs. The model answers questions by grounded selection from its own texts with named sources and declines honestly outside them; it is live at ai.sphuratta.com.
The core is the stable centre of the model, and it is treated as one: nothing is introduced into it after it is built, and it is never modified in place. The model develops by whole new versions of the core, each measured in full before it is adopted; the current version has carried the live service since 16 November 2026.
The model does not invent text. For a question, it finds in its own texts the sentence that answers it and replies with that sentence, naming the source in full. Alongside the answer, the working screen shows the model's reading of the exchange — its own short account of how the answer stands.
If its texts hold no answer, the model says so in plain words and names what its texts actually are. It would rather decline than guess: an honest refusal is a designed behaviour of the model, tested as strictly as its answers.
You can try this now, on the live working screen at ai.sphuratta.com: type a question, dictate it in a browser that supports speech recognition, and listen to the answer read back. The screen carries no description of the model — the address opens the model itself.
The working screen offers three operating modes — A, B and C: three settings of the same core, chosen by the visitor. The screen you land on runs in mode C.
The core is the company's central asset, and it is protected as one:
Today, at the Alpha stage, the model:
The company's applicability research established eight areas for the model. Services are formed in these areas; the plans page of this site lists the services open today.
Where a picture holds together and where it falls apart — the model shows which link brings the whole down.
The model reads a flow's own norm, its deviations, and the early signs that precede a breakdown.
The model finds the nodes everything rests on, and where a step taken in one node will arrive.
The model shows where a body of texts agrees, where it contradicts itself word for word, and where its blind spots are.
Where no single fact is the answer, the model returns a configuration in which the parts agree — or says plainly that no agreeing configuration was found.
A new datum does not get appended to the side: the whole picture is reassembled, and the model shows what shifted.
The model shows a question from every viewpoint at once — including what is not visible from inside a single viewpoint.
The model deepens by cycles of real work; every cycle is reported in numbers, before and after.
Sphuratta AI is at the Alpha stage. The Alpha release was issued on 15 November 2026, after the model passed the stage exam — an hourly testing battery over its answers, its refusals, and the integrity of its core — in full.
What is not in the Alpha, stated plainly:
The stages ahead are Beta, pre-release, and release. A stage is issued only when its exam is passed; there are no announced dates.
Port Alberni, British Columbia, Canada
Contact: project.capital@sphuratta.com
Sphuratta AI is developed and owned by the company. Services built on the model are sold and delivered by the company.