PageIndex Node SDK (pageindex-node-sdk)

Node-first TypeScript SDK for the PageIndex API.
PageIndex is a vectorless, reasoning-based RAG framework that turns long, complex documents into a tree-structured index and lets LLMs perform agentic reasoning over that structure for context‑aware, traceable retrieval — with no vector database and no chunking required.
This SDK wraps the core HTTP endpoints (PDF upload, document status/tree/OCR, listing, deletion, and Chat API with streaming) for easy use from Node.js.

Install

npm install pageindex-node-sdk
# or
yarn add pageindex-node-sdk

Requirements

• Node.js >= 18 (uses built‑in fetch, FormData, ReadableStream)
• A PageIndex API key from the Developer Dashboard

Quickstart (TypeScript / ESM)

import { PageIndexClient } from "pageindex-node-sdk";

const client = new PageIndexClient({
  apiKey: process.env.PAGEINDEX_API_KEY!,
});

async function main() {
  // 1) Upload a PDF
  const { doc_id } = await client.submitDocumentFromPath({
    path: "./2023-annual-report.pdf",
  });
  console.log("doc_id:", doc_id);

  // 2) Poll until tree processing is complete
  let status = "processing";
  while (status !== "completed") {
    const doc = await client.getTree(doc_id);
    status = doc.status;
    console.log("status:", status, "retrieval_ready:", doc.retrieval_ready);
    if (status !== "completed") {
      await new Promise((r) => setTimeout(r, 1500));
    }
  }

  // 3) Non‑streaming chat
  const chat = await client.chatCompletions({
    doc_id,
    messages: [
      { role: "user", content: "What are the key findings in this document?" },
    ],
    enable_citations: true,
  });
  console.log(chat.choices?.[0]?.message?.content);
}

main().catch((err) => {
  console.error(err);
  process.exit(1);
});

Streaming chat example

import { PageIndexClient } from "pageindex-node-sdk";

const client = new PageIndexClient({
  apiKey: process.env.PAGEINDEX_API_KEY!,
});

async function streamChat(docId: string) {
  for await (const evt of client.chatCompletionsStream({
    doc_id: docId,
    messages: [
      { role: "user", content: "Summarize this document in 5 bullet points." },
    ],
    enable_citations: true,
  })) {
    if (evt.type === "data") {
      const delta = evt.data.choices?.[0]?.delta?.content ?? "";
      if (delta) process.stdout.write(delta);
    } else if (evt.type === "done") {
      break;
    }
  }
  process.stdout.write("\n");
}

API overview

All methods live on PageIndexClient:

• Constructor

  • new PageIndexClient({ apiKey, baseUrl?, fetch? })
    • apiKey (required): your PageIndex API key (sent via api_key header).
    • baseUrl (optional): defaults to https://api.pageindex.ai.
    • fetch (optional): custom fetch implementation (e.g. for tests).

• PDF processing

  • submitDocument({ file, filename?, mode? })
    • Wraps POST /doc/ with multipart/form-data.
  • submitDocumentFromPath({ path, filename?, mode? })
    • Node‑only helper that reads a local PDF and calls submitDocument.

• Document results

  • getDocument(docId, { type?, format?, summary? })
    • Generic wrapper over GET /doc/{doc_id}/.
  • getTree(docId, { summary? })
    • Convenience wrapper for type=tree.
  • getOcr(docId, { format? })
    • Convenience wrapper for type=ocr + format=page|node|raw.

• Metadata & management

  • getDocumentMetadata(docId)
    • GET /doc/{doc_id}/metadata
  • listDocuments({ limit?, offset? })
    • GET /docs
  • deleteDocument(docId)
    • DELETE /doc/{doc_id}/

• Chat API (beta)

  • chatCompletions(request)
    • Non‑streaming POST /chat/completions.
  • chatCompletionsStream(request)
    • Streaming via Server‑Sent Events (SSE); yields parsed events with:
      • type: "data" | "done" | "comment"
      • data: parsed chunk (when type === "data").

The request/response shapes are aligned with the official PageIndex API reference.

Contribution Guidelines

• Issues & feature requests

  • Open a GitHub issue with:
    • A clear description of the problem or feature.
    • Steps to reproduce (if applicable).
    • Environment details (Node version, OS).

• Pull requests

  • Fork the repository.
  • Create a feature branch (feat/my-feature or fix/my-bug).
  • Ensure:
    • npm run typecheck passes.
    • npm run lint passes.
  • Write clear commit messages and, where useful, update this README.md or code comments for non-obvious behavior.

• Code style

  • Follow existing patterns in src/pageindex-client.ts and src/types.ts.
  • Prefer small, focused PRs.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Repository

Source code and issue tracking live on GitHub:
https://github.com/coolpinkzz/pageindex-node-sdk