</>MCP Agents Market
Agent

AutoSci

by skyllwt1.6kPythonUpdated 2026-08-19

Karpathy's LLM-Wiki vision, fully realized — wiki-centric full-lifecycle AI research platform powered by Claude Code

Claude CodeCodexOpenCode

AutoSci is a memory-centric AI agent platform that automates the complete scientific research lifecycle from literature review to paper publication. Built for Claude Code and Codex runtimes, it maintains a structured wiki-based knowledge graph across projects and provides 30+ specialized skills covering paper ingestion, idea generation, experiment execution, manuscript drafting, and peer review response. The system implements persistent memory that compounds across research projects, supporting full end-to-end workflows from reading papers through writing rebuttals. AutoSci has been used to produce complete published papers in domains including GPU kernel optimization, biomedical drug discovery, and cognitive AI modeling.

Key Features

Wiki-centric knowledge graph with persistent memory across research projects and automatic cross-reference building
30+ agent skills spanning literature ingestion, idea generation, experiment design/execution, paper writing, and rebuttal generation
Dual-runtime support for both Claude Code (slash commands) and Codex (dollar-sign invocations)
Automated paper discovery and daily arXiv recommendations with optional auto-ingest based on research interests
Multi-model review system using independent LLM as peer reviewer for ideas, experiments, and drafts
Full experiment lifecycle automation including code generation, remote deployment, monitoring, and result collection
LaTeX paper drafting with automatic figure generation, BibTeX verification, and conference poster creation
Bilingual skill support (English/Chinese) with configurable third-party LLM provider compatibility

Use Cases

  • 01Automating literature review by ingesting papers from local PDFs, arXiv URLs, or daily automated feeds into a structured knowledge base
  • 02Generating novel research ideas through multi-phase brainstorming with dual-model validation and pilot experiments
  • 03Designing and executing machine learning experiments with automatic ablation studies, remote GPU deployment, and verdict evaluation
  • 04Drafting complete academic papers from experimental results including Related Work sections, figures, tables, and citations
  • 05Creating conference posters automatically from finished LaTeX manuscripts with extracted figures and tables
  • 06Responding to peer review comments with evidence-mapped rebuttals stress-tested by an independent reviewer LLM

Related Agents

View more

hermes-agent

The agent that grows with you

Agent250.4k

agency-agents

A complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.

Agent155.5k

openinterpreter

A coding agent for open models like Kimi K3

Agent68k

cline

Autonomous coding agent as an SDK, IDE extension, or CLI assistant.

Agent66.3k

AutoSci — FAQ

What is AutoSci and what does it do?+

AutoSci is a memory-centric AI agent system that automates the full scientific research workflow from reading papers to writing rebuttals. It maintains a persistent wiki-based knowledge graph and provides 30+ skills for literature review, ideation, experiments, and paper writing, running on Claude Code or Codex runtimes.

How do I install and set up AutoSci?+

Clone the repository branch for your runtime (main for Claude Code stable, autosci-codex for Codex, autosci-opencode for OpenCode), run the setup script (./setup.sh), configure API keys in .env, add your papers to raw/papers/, then launch your runtime and invoke /init or $init with your research topic. Python 3.9+ and Node.js 18+ are required.

Which AI clients and runtimes does AutoSci work with?+

AutoSci runs on Claude Code (stable main branch) and Codex (autosci-codex branch), with an OpenCode preview (autosci-opencode branch). Claude Code users can use native Anthropic Claude or third-party Anthropic-compatible providers like DeepSeek, Kimi, MiMo, and GLM.

What API keys are required to use AutoSci?+

The agent runtime authentication (claude login for Claude Code or Codex sign-in) is required. Optional keys enhance functionality: Semantic Scholar API for citation graphs, DeepXiv token for semantic search (auto-registered), and LLM_API_KEY + LLM_BASE_URL for cross-model review with any OpenAI-compatible provider.

Is AutoSci free and open source?+

Yes, AutoSci is released under the MIT license and fully open source. However, using it requires API access to Claude Code or Codex, which have their own pricing. The optional review LLM can use free-tier or paid third-party providers.

Can AutoSci run experiments automatically including deployment and monitoring?+

Yes, the /exp-run skill (or $exp-run in Codex) handles the full experiment pipeline including code generation, local or remote GPU deployment via SSH, monitoring with screen sessions, and automatic result collection. The /exp-eval skill then uses an independent reviewer LLM to judge results and update the idea status.

How do I install AutoSci?+

Open the source repository on GitHub and follow its README. AutoSci is a agent — MCP Agents Market links you directly to the official repo.

Is AutoSci free?+

AutoSci is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches

Research AgentsProductivity & Workflow AgentsKnowledge & Memory AgentsBest MCP ServersBest SkillsBest PluginsTop rated listingsClaude Code toolsBrowse all categories