Quickstart

Installation

Install Cell and its peer dependencies:

npm install @claritylabs-inc/cell ai pdf-lib

Then install a model provider:

# Anthropic (default)
npm install @ai-sdk/anthropic

# OpenAI
npm install @ai-sdk/openai

# Google
npm install @ai-sdk/google

Extract a policy

The simplest path — default Anthropic models, no configuration:

import { classifyDocumentType, extractFromPdf, applyExtracted } from "@claritylabs-inc/cell";
import { readFileSync } from "fs";

// Load a PDF as base64
const pdfBase64 = readFileSync("./policy.pdf").toString("base64");

// Step 1: Classify — is this a policy or a quote?
const { documentType, confidence } = await classifyDocumentType(pdfBase64);
console.log(`Classified as ${documentType} (confidence: ${confidence})`);

// Step 2: Extract — run the full multi-pass pipeline
const { extracted } = await extractFromPdf(pdfBase64);

// Step 3: Apply — map raw extraction to structured fields
const fields = applyExtracted(extracted);
console.log(fields.carrier);        // "Hartford"
console.log(fields.policyNumber);   // "GL-2024-001234"
console.log(fields.coverages);      // [{ name: "General Liability", limit: "$1,000,000", ... }]

Extract a quote

Quotes have a separate pipeline that captures quote-specific fields like subjectivities and premium breakdowns:

import { extractQuoteFromPdf, applyExtractedQuote } from "@claritylabs-inc/cell";

const { extracted } = await extractQuoteFromPdf(pdfBase64);
const fields = applyExtractedQuote(extracted);

console.log(fields.quoteNumber);          // "Q-2024-5678"
console.log(fields.premiumBreakdown);     // [{ line: "GL", amount: "$5,200" }, ...]
console.log(fields.subjectivities);       // [{ description: "Loss runs required", ... }]

Use a custom model

Bring any model from any provider:

import { createOpenAI } from "@ai-sdk/openai";
import { extractFromPdf, createUniformModelConfig } from "@claritylabs-inc/cell";

const openai = createOpenAI();
const { extracted } = await extractFromPdf(pdfBase64, {
  models: createUniformModelConfig(openai("gpt-4o")),
  metadataProviderOptions: {},  // disable Anthropic-specific thinking
});

Add logging

Every pipeline function accepts a log callback:

const { extracted } = await extractFromPdf(pdfBase64, {
  log: async (msg) => console.log(`[cell] ${msg}`),
});

Output:

[cell] Pass 1: Extracting metadata...
[cell] Calling model (max 4096 tokens)...
[cell] 12450 in / 2300 out tokens (3.2s)
[cell] Document: 45 page(s)
[cell] Pass 2: Extracting sections pages 1–15...
...

Next steps