Skip to content

Scrinium

Scrinium — a research infrastructure for AI agents (hard fork of ScholarAIO).

This site is published at https://wszqkzqk.github.io/scrinium/ via GitHub Pages.

Scrinium is a research infrastructure for AI agents. You interact with your literature knowledge base through natural language — searching, reading, analyzing, and writing — all from the command line.

Features

  • PDF Ingestion: Convert PDFs to structured Markdown via MinerU (cloud or local)
  • Keyword Search: FTS5 full-text search with field-weighted ranking; the agent extends recall through query expansion, curated tags, and citation-graph snowballing
  • Tag-Based Topics: A controlled tag vocabulary curated by the agent doubles as the topic system — distribution overview and drill-down via scrinium topics
  • Citation Graph: View references, citing papers, and shared references
  • BibTeX Export: Filtered export with standard citation formats
  • Paper Translation: Agent-driven chunked translation stored as paper_{lang}.md, readable via show --lang
  • Literature Exploration: Multi-dimensional OpenAlex queries with isolated data
  • Workspace Management: Organize papers into subsets for focused work
  • Federated Discovery: Search your library, explore silos, and arXiv in one flow
  • Research Insights: Inspect search/read behavior trends
  • Scientific Tool Docs: Query indexed official docs for scientific computing tools with toolref
  • Extensible Tool Onboarding: Keep adding the next scientific tool users need through a documented onboarding workflow
  • Office Document Inspection: Verify DOCX / PPTX / XLSX structure with document inspect
  • Agent Skills: Reusable workflows for search, writing, scientific runtime, and more

Scrinium makes no in-framework LLM or embedding calls: anything that requires understanding (summarizing, translating, classifying, recommending) is done by the agent itself, with the framework providing storage, retrieval, and queue primitives.

Quick Start

pip install "scrinium[full]"
scrinium setup

See Installation for detailed instructions. If you are working from a local clone or contributing to Scrinium itself, use the editable install path shown there instead. See Agent Setup for repo-open vs plugin setup paths. See Translation Guide for the agent-driven translation workflow and storage conventions. See Insights Guide for reading/search behavior analytics. See API Reference for Python module documentation.

Two Usage Modes

Mode Interface Best for
Agent Any AGENTS.md-compatible coding agent Full research workflow via natural language
CLI Terminal Scripting and automation