Repolode
Submit
PKU-DAIR

PKU-DAIR/AutoSci

An agentic AI research system built on Claude Code or Codex that maintains a persistent, cross-project wiki covering the research lifecycle from paper ingestion to rebuttal.

View on GitHub
Stars
1,727
Forks
215
Open issues
4
Last commit
9 days ago

What is it?

AutoSci is an AI research agent that organizes scientific work around a structured, persistent wiki. Papers, notes and web pages are ingested into the wiki, which then supports idea generation, experiment design and execution, paper drafting, poster creation and rebuttal writing. Knowledge accumulates across projects rather than resetting with each session.

The system runs as a set of bilingual agent skills inside Claude Code (main branch) or Codex and OpenCode (preview branches). Commands such as /ingest, /ideate, /exp-design, /exp-run, /paper-draft and /rebuttal each handle one stage. An optional second LLM can act as an independent reviewer for ideas, experiments and drafts through any OpenAI-compatible API.

The project includes a local web UI and Obsidian export for browsing the knowledge graph, a daily arXiv recommendation workflow that can deliver digests by email, and a discovery command that ranks papers from specific conference venues against existing wiki topics.

Key features

  • 01Imports local papers, notes and web pages into a persistent research wiki.
  • 02Covers ideation, experiment design and execution, evaluation, and paper writing in separate skills.
  • 03Supports Claude Code (stable), with Codex and OpenCode preview branches.
  • 04Allows a second OpenAI-compatible LLM to review ideas, experiments and drafts.
  • 05Provides a browser knowledge-graph view and Obsidian canvas export.
  • 06Compiles drafted LaTeX papers into conference posters and PDFs.

Use cases

  • Building a literature wiki for a new research direction from your own paper collection.
  • Turning a research idea into designed, executed and evaluated experiments with an agent.
  • Drafting a LaTeX paper and preparing a conference poster from the same wiki.
  • Preparing structured responses to reviewer comments grounded in your existing notes.

Quick start

  1. 1

    Clone the Codex Preview branch

    git clone -b autosci-codex https://github.com/skyllwt/AutoSci.git
  2. 2

    Enter the project directory

    cd AutoSci
  3. 3

    Run the setup script in English

    ./setup.sh --lang en
  4. 4

    Start Codex in the project, then invoke the init skill with your research topic

    codex

Things to know

  • The project is marked as internal beta, and features are still being tested and iterated.
  • The main Claude Code version is stable, but the Codex and OpenCode versions are only previews with known boundaries.
  • It runs on top of Claude Code or Codex, so you need one of those agent runtimes.

In collections

Claude Code tools, skills & pluginsOpen-source AI agent frameworks & apps

Topics

#ai-agent#research-automation#llm#claude-code#knowledge-base#literature-review#paper-writing#arxiv

FAQ

What is AutoSci?
An agentic AI research system built on Claude Code or Codex that maintains a persistent, cross-project wiki covering the research lifecycle from paper ingestion to rebuttal. AutoSci is an AI research agent that organizes scientific work around a structured, persistent wiki. Papers, notes and web pages are ingested into the wiki, which then supports idea generation, experiment design and execution, paper drafting, poster creation and rebuttal writing. Knowledge accumulates across projects rather than resetting with each session.
Is AutoSci free to use? What license does it have?
AutoSci is open source under the MIT license. Check the license terms before using it in commercial products.
Is AutoSci still maintained?
Actively maintained. The last commit to AutoSci was on Oct 30, 2026.
What can I use AutoSci for?
Building a literature wiki for a new research direction from your own paper collection. Turning a research idea into designed, executed and evaluated experiments with an agent. Drafting a LaTeX paper and preparing a conference poster from the same wiki. Preparing structured responses to reviewer comments grounded in your existing notes.
How do I install AutoSci?
Clone the Codex Preview branch: git clone -b autosci-codex https://github.com/skyllwt/AutoSci.git Enter the project directory: cd AutoSci Run the setup script in English: ./setup.sh --lang en Start Codex in the project, then invoke the init skill with your research topic: codex
Where can I get AutoSci?
The source code and installation instructions are on GitHub: https://github.com/PKU-DAIR/AutoSci

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects