PDF Q&A

npm install @agentskit/runtime @agentskit/adapters @agentskit/rag @agentskit/memory pdf-parse

import { createRuntime } from '@agentskit/runtime'
import { openai, openaiEmbedder } from '@agentskit/adapters'
import { createRAG } from '@agentskit/rag'
import { fileVectorMemory } from '@agentskit/memory'
import { readFileSync } from 'node:fs'
import pdfParse from 'pdf-parse'

const [pdfPath, ...questionParts] = process.argv.slice(2)
const question = questionParts.join(' ')

if (!pdfPath || !question) {
  console.error('Usage: tsx ask-pdf.ts <file.pdf> "<question>"')
  process.exit(1)
}

// 1. Extract text
const data = await pdfParse(readFileSync(pdfPath))
const fullText = data.text

// 2. Chunk (simple: paragraphs of 500-ish chars)
function chunk(text: string, size = 500): string[] {
  const paragraphs = text.split(/\n\n+/)
  const chunks: string[] = []
  let current = ''
  for (const p of paragraphs) {
    if ((current + p).length > size) {
      if (current) chunks.push(current)
      current = p
    } else {
      current += '\n\n' + p
    }
  }
  if (current) chunks.push(current)
  return chunks
}

// 3. Index in-memory (per-PDF, ephemeral)
const rag = createRAG({
  store: fileVectorMemory({ path: `./.cache/${pdfPath}.embeddings.json` }),
  embed: openaiEmbedder({ apiKey: KEY, model: 'text-embedding-3-small' }),
  topK: 4,
})

await rag.ingest(
  chunk(fullText).map((content, i) => ({
    id: `chunk-${i}`,
    content,
    source: `${pdfPath}#${i}`,
  })),
)

// 4. Ask
const runtime = createRuntime({
  adapter: openai({ apiKey: KEY, model: 'gpt-4o-mini' }),
  retriever: rag,
  systemPrompt:
    'Answer using only the provided document excerpts. ' +
    'Cite passages by their source index. Say "not found in the document" if absent.',
})

const result = await runtime.run(question)
console.log(result.content)

npx tsx ask-pdf.ts paper.pdf "What is the main contribution of this work?"