@noctara/sdk

The only identity layer that travels.

Behavioral identity primitive for AI surfaces. One import. Your agent reads the person it is talking to, in their rhythm, with their compliance lever, without ever asking for chat history.

Install

npm install @noctara/sdk

Use

import { Noctara } from "@noctara/sdk";
import Anthropic from "@anthropic-ai/sdk";

const noctara = new Noctara();
const claude = new Anthropic();

const { system_prompt } = await noctara.systemPrompt({
  email: "[email protected]",
});

const msg = await claude.messages.create({
  model: "claude-opus-4-7",
  max_tokens: 1024,
  system: system_prompt + "\n\n# Your task.\nYou are this person's agent.",
  messages: [{ role: "user", content: "What should I do tomorrow?" }],
});

API

• noctara.systemPrompt({ email | mark }, { style }) returns a Claude-ready system addendum tuned to the user's compressed identity.
• noctara.identityToken({ email | mark }, { audience, ttlSeconds }) mints a signed JWT carrying word, rhythm, force.
• noctara.verifyToken(token) verifies an identity token and returns the claims.
• noctara.scrub(text) strips em-dashes, en-dashes, smart quotes, and the most reliable AI tells.

Voice rules

The system prompt returned by Noctara carries hard voice rules. Models that respect them sound like a friend who has been listening for years. Models that ignore them sound like LLMs. The choice is yours.

• No em-dashes. No en-dashes. Periods or new sentences.
• No AI tells.
• Banned words: resilient, authentic, journey, manifest, vibe, energy, passion.
• Last sentence of any response is the most compressed.

Patent

The primitive is covered by US provisional patent #64/048,624. Non-commercial use is free. Commercial use up to 10,000 MAU is free under attribution. Above that, contact [email protected].

License

MIT for this SDK. The behavioral identity primitive itself is patent-protected. See noctaracorp.com/spec for the full specification.