Drop-in utilities for the Hebrew agent stack.

Link AI publishes a small set of focused libraries and curates a larger one. The official SDK is incubating behind the private preview; the open-source utilities below are production-ready today and stand on their own — use them with or without a Link AI workspace.

Official packages

hebrew-text-utils (npm)

Battle-tested helpers for normalising, tokenising, and nikud-stripping Hebrew text before you feed it to a model or an agent prompt. Ships as ESM + CJS, zero runtime dependencies, and is what powers Link AI Voice’s Hebrew transcript post-processing.

npm install hebrew-text-utils
# or
bun add hebrew-text-utils

import { stripNikud, normalize, isHebrew } from "hebrew-text-utils";

stripNikud("שָׁלוֹם");        // "שלום"
normalize("שלום   עולם");     // "שלום עולם"
isHebrew("Hello עולם");      // true (mixed-script detection)

• npm: npmjs.com/package/hebrew-text-utils
• source: github.com/linkaiil1234/hebrew-text-utils
• licence: MIT.

@linkai/sdk (private preview)

Typed JS/TS client for the REST API. The same client used by Link AI internally for trigger.dev workflows. The package installs from a private npm scope while the API is in design- partner preview. Email office@linkaiil.com to request access.

import { LinkAI } from "@linkai/sdk";

const linkai = new LinkAI({ apiKey: process.env.LINKAI_TOKEN! });

// Typed surface mirrors the REST endpoints.
const agents = await linkai.voice.agents.list();
const call = await linkai.voice.calls.create({
  agentId: agents[0].id,
  to: "+972541234567",
});

Curated: awesome-hebrew-ai-voice

A hand-maintained directory of every model, dataset, and tool worth knowing if you build Hebrew-native voice or text agents. Contributions welcome — Hebrew NLP is small enough that one good link genuinely moves the field.

• repo: github.com/linkaiil1234/awesome-hebrew-ai-voice
• sections: ASR models, TTS voices, evaluation corpora, tokenizers, annotation tooling, latency benchmarks.
• updated continuously; PRs reviewed within a week.

Community wrappers

The REST API is plain JSON, so any HTTP client works. A handful of community libraries add ergonomic types or framework helpers on top.

• Python. Use any HTTP client (httpx, requests). A typed Python SDK is on the roadmap; share a vote at office@linkaiil.com to nudge it forward.
• Go. Community library linkai-go tracks the public endpoint surface — see the awesome list for the current maintainer.
• n8n. A community node ships a workspace-aware credential helper and pre-built nodes for the most common flows. Search “Link AI” in the n8n community registry.

Zero-dependency calls

If you would rather skip every library, the REST API is plain enough that a single fetch wrapper does the job. The snippet below is what we recommend for serverless environments where cold-start size matters.

type LinkAIInit = {
  apiKey: string;
  baseUrl?: string;
};

export function linkai({ apiKey, baseUrl = "https://api.linkaiil.com/v1" }: LinkAIInit) {
  return async function call<T>(path: string, init: RequestInit = {}): Promise<T> {
    const res = await fetch(`${baseUrl}${path}`, {
      ...init,
      headers: {
        Authorization: `Bearer ${apiKey}`,
        "Content-Type": "application/json",
        ...init.headers,
      },
    });
    if (!res.ok) throw new Error(`Link AI ${res.status} on ${path}`);
    return res.json() as Promise<T>;
  };
}

// Usage:
// const api = linkai({ apiKey: process.env.LINKAI_TOKEN! });
// const me = await api<{ workspace_id: string }>("/whoami");