Xueyang-Song/paper-pilot

Research-paper workflow tool for literature triage and structured reading

Paper Pilot – A Local‑First AI Research Assistant

What it is – Paper Pilot is a desktop application (Electron + React) that helps scientists collect, store, and synthesize academic papers. It crawls multiple open‑access scholarly databases, downloads PDFs, converts them to plain‑text, indexes the content with full‑text and vector search, and lets you ask natural‑language questions that are answered by a local or hosted LLM (e.g., Ollama, OpenAI‑compatible APIs). All data stays on your machine unless you deliberately send a request to a remote model.

Core capabilities

Capability How it works
Multi‑source crawling Simultaneously queries 8+ sources (OpenAlex, Crossref, Semantic Scholar, PubMed/PMC, arXiv, Europe PMC, CORE, Unpaywall, and an experimental Google Scholar scraper).
PDF → knowledge pipeline Retrieves open‑access PDFs via Unpaywall, converts them to Markdown with the MarkItDown tool, and stores the text in a SQLite database.
Local storage & search SQLite with the built‑in FTS5 engine for keyword search and the sqlite‑vec extension for vector similarity search. Projects are isolated so you can keep separate research contexts.
AI‑assisted synthesis Sends a user’s query plus the relevant retrieved passages to a local Ollama model or any OpenAI‑compatible endpoint. Supports streaming chats, “grounded” mode (only returns information that can be traced to a source) and “exploratory” mode (allows more speculative answers).
Auditable evidence review Every answer is linked to the exact citation(s) used; the UI shows a trace of the reasoning, lets you accept/reject AI suggestions, and exports a reproducible evidence package (RIS, BibTeX, CSV).
Secure credentials API keys are stored using Electron’s safeStorage API, keeping them encrypted on disk.

Architecture at a glance

paper‑pilot/
├─ src/electron/          # Node.js main process
│   ├─ crawlers/          # adapters for each scholarly API
│   ├─ db/                # SQLite + FTS5 + sqlite‑vec layer
│   ├─ agent/             # LLM orchestration (tool‑calling, streaming)
│   └─ python/            # MarkItDown conversion & Playwright bridge
├─ src/renderer/          # React 19 UI (Vite, Tailwind, TanStack Query)
│   ├─ workspace/         # Chat interface
│   ├─ projects/          # Project management
│   ├─ artifacts/         # Paper list & PDF viewer
│   └─ settings/          # API keys, policies, model choice
└─ tests/                 # Vitest test suite

The stack is entirely open‑source: TypeScript 5, React 19, Vite, SQLite, and a thin Python environment for PDF processing.

Getting started (quick‑run)

# 1. Clone
git clone https://github.com/Xueyang-Song/paper-pilot.git
cd paper-pilot

# 2. Install Node dependencies
npm install

# 3. Run the development server (UI opens at http://127.0.0.1:5173)
npm run dev

# 4. Build a production bundle
npm run build

# 5. Package as a native desktop app (macOS/Linux – still untested)
npm run package

Prerequisites: Node ≥ 22.18, Python 3.11+, and optionally Ollama if you want fully offline LLM inference.

Who might benefit?

  • Researchers who need to pull literature from many repositories without juggling dozens of browser tabs.
  • Literature reviewers who want a reproducible audit trail linking every claim to a specific paper.
  • Teams with strict data‑privacy policies because all PDFs and embeddings stay on the local disk.
  • Anyone experimenting with local LLMs (Ollama) for scholarly Q&A.

Current maturity

  • Core crawlers, SQLite indexing, and the AI chat workspace are marked stable.
  • Google‑Scholar scraper is experimental.
  • macOS/Linux packaging has not been fully tested; cloud‑sync collaboration is not planned.

Contributing

The repo includes a full Vitest test suite, linting, and type‑checking scripts. Contributions are encouraged; the maintainers ask that you open an issue before large PRs. See CONTRIBUTING.md for the protected‑branch workflow and release labeling.

TL;DR

Paper Pilot bundles together scholarly‑source crawling, local PDF ingestion, searchable SQLite storage, and an LLM‑driven chat interface—all in a single desktop app that keeps your data private. It’s a practical tool for scientists who want AI assistance without moving their research corpus to the cloud.

Related

  • Project
    usnistgov/dioptra

    Dioptra is NIST’s open‑source Docker‑based platform for testing AI models against trustworthiness criteria (fairness, safety, explainability, etc.). It offers a REST API, web UI, and Python client, lets you define experiments in YAML, supports plugins for PyTorch/TensorFlow, and tracks reproducible, traceable runs for developers, auditors, researchers, and red‑teamers.

  • Project
    NaomiProject/Naomi

    Naomi is an open‑source, always‑on voice‑assistant platform (Python, Linux/Raspberry Pi) that lets you control home devices, query information, and add custom “skills” via simple Python modules.

  • Project
    rossoctl/rossoctl

    Rossoctl is an open‑source, Kubernetes‑native platform that intercepts AI agents via the RossoCortex data‑plane, enforcing identity, authorization, resilience and observability. It adds reusable services (skills, tools, memory, knowledge base, sandboxes) and admin tooling, enabling trustworthy, production‑grade agent deployments across frameworks.

  • Project
    Prorise-cool/Claude-Code-Multi-Agent

    Claude Code Multi‑Agent is an open‑source framework that gives Anthropic’s Claude Code project awareness and expert‑agent capabilities. It uses a Python Hook system and a locally‑run Ollama model to automatically detect the codebase type, analyse user intent, load 300+ domain‑specific “Skills” (backend, frontend, testing, security, etc.), and enforce automatic documentation (DEVELOPMENT.md, KNOWLEDGE.md, CHANGELOG.md). After cloning, installing Ollama, a model, and uv, you drop your project into the repo and Claude Code instantly knows its context, suggests appropriate tools, and can be driven with slash commands like `/backend‑specialist` or workflow commands such as `/kiro/spec`. Key points: - Zero‑config start; all configuration lives in `.claude/settings.json` and `prompts.json`. - Hooks run on session start, user prompt, tool use, etc., handling project detection, intent analysis, and doc updates. - Skills are defined by `SKILL.md` files and are auto‑discovered. - Works entirely locally (no external API), integrates with Git, and is MIT‑licensed. Ideal for developers or small teams who want a reproducible, on‑premise AI coding assistant.