Skip to main content

Instruct-it

Structured information extraction adhering to custom templates and schemas using the Knowledgator TypeScript SDK.


Example Usage​

Structured Output with JSON Template​

import { createClient } from '@knowledgator/typescript-sdk';

const kg = createClient();

const template = {
summary: "",
key_points: [],
risk_level: ""
};

const result = await kg.instruct.it({
text: "Analyze this security policy and provide structured feedback.",
template, // Object or JSON-stringified schema
output_type: "json",
adapter_id: "your-lora-adapter-id" // Optional LoRA adapter
});

console.log(result.outputs.summary);
console.log(result.outputs.key_points);
LoRA Adapters vs. Open Models

adapter_id is for fine-tuned LoRA adapters attached to base services. If you trained a standalone Open Model, query it via kg.model(model_id).predict(...) instead.

With File Content (Upload)​

import { createClient } from '@knowledgator/typescript-sdk';
import { readFile } from 'fs/promises';

const kg = createClient();

const fileContent = await readFile('./document.pdf');

const result = await kg.instruct.it({
text: "Summarize this document.",
files: [
{
name: "document.pdf",
content: new Uint8Array(fileContent)
}
],
output_type: "json"
});

console.log(result.outputs);

Expected Output​

{
"summary": "This document outlines cybersecurity protocols for enterprise data access.",
"key_points": [
"Zero-trust network architecture required for all remote staff",
"Encryption enforced for data at rest and in transit",
"Quarterly access audits mandated"
],
"risk_level": "medium"
}

Result Object Structure​

result.outputs contains the structured response conforming to the supplied template:

// Accessing fields directly
const summary: string = result.outputs.summary;
const keyPoints: string[] = result.outputs.key_points;