pi-literature-review

Local, open, login-free academic literature tool for Pi or the browser: search paper databases plus code repositories, citation graphs, PRISMA diagrams, answers over your PDFs with page-exact citations. Optimised for local AI models.

Packages

Package details

extension

Install pi-literature-review from npm and Pi will load the resources declared by the package manifest.

npm repohome report
$ pi install npm:pi-literature-review
Package
pi-literature-review
Version
0.2.1
Published
Nov 11, 2026
Downloads
396/mo · 253/wk
Author
flindenberger
License
MIT
Types
extension
Size
1.8 MB
Dependencies
2 dependencies · 4 peers
Pi manifest JSON
{
  "image": "https://raw.githubusercontent.com/flindenberger/pi-literature-review/main/docs/img/gallery.png",
  "extensions": [
    "./index.ts"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

 

A local, open, login-free literature tool as an extension for the Pi coding agent or as a standalone browser app: search four academic paper databases plus code repositories, explore citation graphs and PRISMA diagrams, then ask your PDFs and get answers with page-exact citations, optimised for your local AI models.

 

 

Features

  • Search papers and download them directly — arXiv, CrossRef, OpenAlex and Semantic Scholar, verified DOIs, open-access PDFs into a folder you choose
  • Search papers with code repositories — code and data links on every result, plus papers found through GitHub, Hugging Face and awesome lists
  • Search for specific authors and their work — narrow a topic to chosen authors, or list everything they published, optionally as first author
  • Ask questions and write summaries or synthesis reports about your PDF library — page-exact citations, highlighted where they come from
  • Drag and drop your PDFs — straight into the browser and start asking
  • Optimised for local AI models — tested with 8–9B models on an 8 GB graphics card; search needs no model at all
  • Citation graphs — see what a paper builds on and who cites it
  • PRISMA diagrams — a reproducible search record, from query to final list
  • No language model ever touches paper data — titles, authors, DOIs, abstracts and citations come from the source APIs, checked and inserted by fixed code

Install

As Pi coding agent extension

pi install npm:pi-literature-review

In Pi, call /pi-literature-review for the browser interface. For the terminal version, call /lit-search or /lit-synthesis.

As standalone

To use the standalone version in your browser, run:

npx pi-literature-review web

Needs Node.js 22.19 or newer. Searching needs no model. Chat and reports use the model selected in Pi, or Ollama, llama.cpp or any OpenAI-compatible server, plus a small embedding model the package fetches for you — see Requirements.

Privacy

No account, no telemetry. Search terms and paper identifiers go to the public scholarly APIs. PDF text goes only to your own embedding and chat models; with a remote model, excerpts go to that provider.

Documentation

Requirements · Search & selection · Summary & synthesis · Browser · Configuration · Command line · Development

How to cite

If the package supports your work, please cite it:

Lindenberger, F. (2026). pi-literature-review (Version 0.2.0) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.22954995

The DOI always resolves to the latest version. BibTeX and other formats: CITATION.cff or "Cite this repository" in the sidebar of this page.

Contributing

Questions, bug reports and pull requests are welcome — open an issue; CONTRIBUTING.md says what helps. Please report security problems privately, as described in SECURITY.md.

About

Developed in the research project Satellite Remote Sensing and Field Data Fusion for Hydraulic Engineering (BAW, UPC, TU Dresden). The multi-source search was inspired by paper-search-mcp (MIT). The synthesis stage can use OpenScholar-8B (Asai et al., 2024, arXiv:2411.14199) as its writing model; the package was first built around it.

Use of AI. Developed using Claude Code as a development tool (code, tests, documentation). The maintainer made the design decisions and tested the changes.

License. MIT — see LICENSE.