RAGOpen SourceFreeActiveLocal hardware· intermediate · ~30 min setup

Vectorless RAG with PageIndex

Build high-accuracy RAG without embeddings, chunking, or a vector DB.

by Shilpa Mitra· verified 1mo ago· v1.0.0

Run this workflow

See exactly what it produces before you build it.

Build this with your agent

One copy-paste hands Claude Code, Codex, or Cursor the full recipe, steps included, nothing to fetch.

Intended Use

Single-document or small-corpus QA where you want strong accuracy without the ops complexity of embeddings + a vector DB.

Not for

  • Corpora over ~10K documents (no index sharding)
  • Sub-second retrieval requirements

The Stack

DeepSeek V4modelOllamaruntimePageIndexretrieval

Tested Against

pageindex@0.4ollama@0.5deepseek-v4

Side effects & data flow

Network
none, local only
Writes
./pageindex_cache/
Credentials
none required

Steps

  1. 1

    Install PageIndex

    Install the PageIndex library.

    pip install pageindex
  2. 2

    Build the index

    PageIndex parses your PDF into a tree.

    from pageindex import PageIndex
idx = PageIndex.from_pdf('./corpus.pdf')
idx.save('./pageindex_cache/')
  3. 3

    Query

    Ask a question via the index.

    from pageindex import PageIndex
idx = PageIndex.load('./pageindex_cache/')
print(idx.query('What was Q3 revenue?', model='ollama/deepseek-v4'))

Eval, 1 fixture

Last passed: verified 1mo ago
  • financebench-q3rubrictimeout 120s · max $0.1

    Judge: claude-sonnet-4-5 Rubric: PASS if (1) the answer contains a specific dollar figure, (2) it cites a page or section, and (3) the figure matches the ground truth ±0.1B.

Results

98.7% on FinanceBench in the original benchmark.

Did this work for you?

Our CI checks the setup runs. You tell us if the whole thing worked. Tell us straight.

Related workflows

Liked this workflow?

Get new verified workflows in WebAfterAI, three issues a week (Tue, Thu, Sat).