Developers

Three lines of code. Every regulated market open.

Your product runs customer data through a public model, and your regulated buyers need proof that the model never sees it. Nakato gives you that proof as an SDK: the model only ever meets the twin, and your product still returns the real answer.

The RSP SDK is in development. Early access is by registration.

integration.ts@nakato/sdk

1const rsp = new Nakato({ apiKey: process.env.NAKATO_KEY });2const { transformed, session } = await rsp.transform(transcript, systemPrompt);3const finalOutput = await rsp.reverse(response.text, session);

The marked value is the twin. It is the only version of your customer the model ever meets.

The problem

The enterprise wall.

Your product works and your customers rely on it. Then a bank, an insurer or a healthcare provider asks one question.

“Where does our data go?”

Asked in every regulated buyer’s security review.

Today

The honest answer is that it goes to a public model provider for inference. Certifications, encryption in transit and no-training promises are all worth having, and none of them changes that answer: the customer’s data still reaches the model in plaintext.

With Nakato built in

Nakato RSP gives you a different answer to give:

“Your real data never reaches the model. The model only ever meets its twin.”

Over half of organisations name data privacy as the top reason they have not adopted AI. Your product is not the problem. The trust gap is. Once it is built into your product, the banks, insurers and healthcare providers that could not say yes become a market you can sell into.

Integration

RSP as an integration layer.

The Nakato SDK sits between your application and the model. Before anything reaches it, RSP finds each sensitive entity (names, addresses, account numbers, financial and medical detail, commercial terms) and replaces it with a semantic twin: fictional but faithful. The model processes fiction. Your product returns facts.

The swap is fully reversible. Your users see their real data in the final output. The model only ever meets the twin.

Before

integration.ts

// Before: your current LLM callconst response = await llm.complete({  model: "claude-sonnet-5",  messages: [{ role: "user", content: transcript }]});

With Nakato

integration.ts

// After: with Nakatoimport { Nakato } from "@nakato/sdk";const rsp = new Nakato({ apiKey: process.env.NAKATO_KEY }); const { transformed, session } = await rsp.transform(  transcript,  systemPrompt); const response = await llm.complete({  model: "claude-sonnet-5",  messages: [    { role: "system", content: systemPrompt },    { role: "user", content: transformed }  ]}); const finalOutput = await rsp.reverse(response.text, session);

Ruled lines are the ones you add. The call to the model is the one you already have; it now carries the twin.

Construct. One client, made once, with your key. The two calls that follow are made on it, and nothing else in your code has to know it is there.

Transform. Reads the task from the system prompt, finds each sensitive value and swaps it for its semantic twin. This is the only version the model meets.

Reverse. Takes the model’s answer and the session, and puts every real value back where its twin stood. Your users read facts.

The system prompt tells RSP what the model has been asked to do, so it knows what to preserve and what to twin. Summarising a call keeps one set of entities; calculating an exposure keeps another.

What it unlocks

What your customers get.

With RSP built in, your product goes from “not approved for regulated data” to “approved and auditable”.

Regulated market access
Your product becomes deployable in financial services, insurance, healthcare, legal and government: the markets that turn away AI products which send customer data to an external model. With RSP in your stack, the model only ever meets the twin.
Enterprise procurement access
Pass the security reviews, DPIAs and vendor risk assessments your deals are waiting on. RSP supplies the audit trail and the data-governance evidence a procurement team asks for, so the reviewer has the proof in front of them.
Coming soon
Verifiable trust mark
Independent verification that a product twins sensitive data before AI inference, so a customer’s real data never reaches the model. Think of it as the SOC 2 of AI data handling.

Architecture

An inline layer, inside your boundary.

Nakato runs as an inline layer in your own stack. Your application sends text to the SDK, which detects the sensitive entities, swaps each one for a semantic twin and hands the twinned text on for inference. When the answer comes back, the SDK reverses the swap, mapping every fictional entity back to its original. Session state is ephemeral and the SDK scales horizontally.

Your application sends the text as it is, and the real value travels only as far as the SDK.
The SDK swaps the value for its semantic twin, and only the twin crosses your boundary.
The model answers about the twin, the only version of the value it ever meets.
The SDK reverses the swap, and your application receives the answer with the original restored.

Real text travels one leg only, between your application and the SDK, and both sit inside your environment. What crosses the boundary, in either direction, is the twin.

State: Processing is scoped to the session, and the state it keeps is ephemeral.
Models: Model-agnostic. It works with any LLM API.
Deployment: Designed to deploy on-premises, including in air-gapped environments.
Uncertainty: Fail-closed: an entity the detector is not certain about is held back, never passed through.
Audit: A full audit trail, with SHA-256 verification for every session.
Packaging: Designed to ship as an npm package, a Python package or a REST API.

Assurance and deployment particulars

Build with us.

The RSP SDK is in development. We are looking for design partners: AI product companies that want to open regulated enterprise markets, and are willing to integrate early, tell us what is missing, and shape the API surface while it can still be shaped.

Design partners receive

Priority access to the SDK beta
Direct engineering support during integration
Co-marketing as a launch partner
Preferential pricing, locked for 24 months