incite-app 0.1.3

Local-first citation recommendation system

inCite

Write text. Get relevant papers from your library.

Overview

inCite is a local-first citation recommendation system. It indexes your Zotero library or a folder of PDFs and suggests relevant papers as you write. Everything runs on your machine -- no cloud account, no API keys, no data leaving your laptop.

• Local-first: Your papers and writing stay on your machine
• Works with what you have: Zotero library, a folder of PDFs, or a JSONL corpus
• Editor plugins: Obsidian, VS Code, Google Docs, and Microsoft Word
• Fine-tuned models: Citation-specific sentence transformers trained on 64K academic citation contexts

Quick Start

pip install incite
incite setup

The setup wizard auto-detects your Zotero library (or accepts a folder of PDFs), builds a search index, and verifies everything works.

Usage

Command Line

# Get recommendations for a passage
incite recommend "The relationship between CO2 emissions and global temperature..." -k 10

# Start the API server (for editor plugins)
incite serve --embedder minilm-ft

# Start the menu bar app (macOS, manages the server for you)
pip install incite[tray]
incite tray

Python API

from incite.agent import InCiteAgent

# From Zotero library
agent = InCiteAgent.from_zotero(embedder_type="minilm-ft")

# From a folder of PDFs
agent = InCiteAgent.from_folder("~/Papers")

# Get recommendations
response = agent.recommend("climate change and agricultural productivity", k=10)
for rec in response.recommendations:
    print(f"  {rec.rank}. [{rec.score:.2f}] {rec.title} ({rec.year})")

REST API

incite serve --embedder minilm-ft
# API docs at http://localhost:8230/docs

curl -X POST http://localhost:8230/recommend \
  -H "Content-Type: application/json" \
  -d '{"query": "climate change impacts on crop yields", "k": 5}'

Editor Plugins

inCite integrates with your writing environment via editor plugins that connect to the local API server.

Editor	Status	Install
Obsidian	Stable	Build from editor-plugins/obsidian-incite/
VS Code	Stable	Build from editor-plugins/vscode-incite/
Google Docs	Stable	Apps Script add-on via clasp push
Microsoft Word	Beta	Office.js add-in, sideload manifest.xml

All plugins share the @incite/shared TypeScript package for API communication and context extraction.

Paper Sources

• Zotero (recommended): Auto-detects your local Zotero library and reads directly from the SQLite database
• PDF folder: Point at any directory of PDFs -- metadata is extracted automatically
• JSONL corpus: Load a pre-built corpus file with title, abstract, authors, and other metadata

How It Works

1. Embed papers: Each paper is embedded as title. authors. year. journal. abstract using a sentence transformer
2. Embed your writing: Your text is embedded with the same model
3. Search: FAISS finds the nearest papers by cosine similarity
4. Fuse (optional): BM25 keyword matching is combined with neural results via Reciprocal Rank Fusion for improved recall
5. Evidence: The best matching paragraph from each paper's full text is attached as supporting evidence

Embedder Models

Model	Key	Dims	Notes
MiniLM fine-tuned v4	minilm-ft	384	Default. Citation-specific, auto-downloads from HuggingFace
MiniLM	minilm	384	Fast, good baseline
SPECTER2	specter	768	Scientific domain
Nomic v1.5	nomic	768	Long context (8K tokens)
Granite	granite	384	IBM Granite, 8K context

For even better results (MRR 0.550 vs 0.428), try the cloud service at inciteref.com which uses our best fine-tuned model.

Development

git clone https://github.com/galenphall/incite.git
pip install -e ".[dev]"
pytest
ruff check src/incite && ruff format src/incite

Optional Dependencies

inCite's core is Apache 2.0 licensed. Some optional features depend on copyleft-licensed libraries and are packaged as extras to keep the default installation permissive.

pip install incite[pdf]       # PyMuPDF for PDF text extraction (AGPL)
pip install incite[zotero]    # pyzotero for Zotero integration (GPL)
pip install incite[api]       # FastAPI server
pip install incite[webapp]    # Streamlit UI
pip install incite[tray]      # macOS menu bar app
pip install incite[all]       # Everything

Note: The pdf and zotero extras pull in AGPL and GPL dependencies respectively. If license compatibility matters for your use case, install only the extras you need.

Cloud Service

inciteref.com offers a hosted version of inCite with additional features:

• Better model: Granite-FT fine-tuned embedder (MRR 0.550 vs 0.428 for the default local model)
• Cloud PDF processing: Full-text extraction without running GROBID locally
• Reference manager: Collections, tags, notes, and citation export (BibTeX/RIS)
• Multi-device sync: Access your library from anywhere

The local CLI and cloud service are complementary -- use whichever fits your workflow.

Contributing

Contributions are welcome. See CONTRIBUTING.md for guidelines.

License

Apache License 2.0

Citation

@software{incite2025,
  author       = {Hall, Galen},
  title        = {inCite: Local-First Citation Recommendation},
  year         = {2025},
  url          = {https://github.com/galenphall/incite},
  license      = {Apache-2.0}
}