openresearch:基于知识图谱的 AI/ML 研究协作系统项目

用户可借助该项目将研究组织为动态知识图谱,支持人机协作开展可追踪、可验证的 AI/ML 研究。核心功能包括声明-证据原子拆解、自动验证计划生成、实验结果结构化回写及图谱持续演化,实现研究过程的透明化与长期积累。【此简介由AI生成】

Branch64Tags4
FilesLast commitLast update
8 months ago
8 months ago
8 months ago
8 months ago
7 months ago
6 months ago
8 months ago
8 months ago
8 months ago
5 months ago
7 months ago
8 months ago
8 months ago
8 months ago
8 months ago
8 months ago
8 months ago
7 months ago
8 months ago
7 months ago
7 months ago
7 months ago
8 months ago
6 months ago
8 months ago
8 months ago
8 months ago
7 months ago
7 months ago
6 months ago
7 months ago
8 months ago
8 months ago
8 months ago
8 months ago

OpenResearch

OpenResearch treats research as an evolving graph of claims, evidence, and decisions — not a linear pipeline from idea to paper.

OpenResearch is an open-source AI/ML research system for structured, traceable, human-AI collaboration.

Its core idea is simple:

  • Research should be decomposed into atomic scientific units
  • Each unit is represented as Claim + Evidence
  • The whole project is maintained as an evolving atom knowledge graph
  • AI operates around claim verification: planning experiments, writing code, running evaluations, and attaching results back to the graph
  • Humans stay in control of problem selection, judgment, and decision-making

Instead of treating research as scattered chats, temporary scripts, and final papers, OpenResearch turns it into a persistent, inspectable, and continuously evolving research state.

research_loop.png

Quick Start


Research Loop

OpenResearch is built around a closed research loop:

  1. A human or AI proposes a new claim
  2. AI expands the local graph context
  3. AI generates a verification plan
  4. AI writes code and runs experiments
  5. Results are attached back as structured evidence
  6. Humans accept, reject, refine, or split claims
  7. The graph evolves
  8. New claims emerge from the updated graph

This makes research a continuous graph expansion process, rather than a one-shot workflow.


Core Abstraction: Atom Knowledge Graph

The system models research as a graph of scientific atoms.

Each atom contains:

  • Claim — a precise scientific statement
  • Evidence — derivations, experiments, observations, or results supporting or challenging the claim

Atoms can take different roles, such as:

  • Fact
  • Method
  • Theorem
  • Verification

Atoms are connected by typed relations, such as:

  • motivates
  • formalizes
  • derives
  • analyzes
  • supports
  • contradicts
  • verifies

This structure allows OpenResearch to represent not only final conclusions, but also the reasoning path, intermediate decisions, failed attempts, and unresolved contradictions behind them.


Why Graph Instead of Paper

A paper is a compressed final artifact.
It usually hides the parts that matter most during actual discovery:

  • abandoned branches
  • intermediate claims
  • failed experiments
  • fragile assumptions
  • alternative explanations
  • unresolved contradictions

OpenResearch is graph-first, not paper-first.

The goal is to preserve the actual research state as it evolves, so future work can build on what was tried, what failed, what was learned, and why decisions were made.


How OpenResearch Differs from Existing AI Research Pipelines

Many AI research systems focus on end-to-end automation: from idea generation to experiment execution to paper writing.

OpenResearch focuses on something different:

  • atomic-level traceability
  • inspectable reasoning state
  • persistent research memory
  • human-in-the-loop control
  • fine-grained scientific collaboration

The goal is not just to produce papers faster.
The goal is to build a system where research remains structured, reviewable, and reusable over time.


What OpenResearch Enables

OpenResearch is designed to support workflows such as:

  • parsing papers into structured claim–evidence atoms
  • building and maintaining atom knowledge graphs
  • proposing and refining new claims
  • generating executable verification plans
  • writing and running experiment code
  • collecting results as structured evidence
  • attaching evidence back to the graph
  • reviewing claims through accept / reject / refine / split decisions
  • managing long-term research memory across iterations
  • enabling collaboration between humans and AI on the level of individual claims and experiments

Current Focus

OpenResearch is currently focused on AI/ML research workflows, especially those that benefit from:

  • structured literature understanding
  • claim-level experiment design
  • experiment tracking and result aggregation
  • persistent project memory
  • iterative human-AI collaboration

Project Status

OpenResearch is still in an early stage and under active development.

The current system already explores a full research loop:

  • project initialization from papers
  • atom graph construction
  • experiment creation and execution
  • result summarization
  • evidence assessment and graph update

Vision

OpenResearch is not just an “AI scientist.”
It is a research operating system.

A system where:

  • ideas become claims
  • claims become executable validations
  • results become structured evidence
  • contradictions remain visible
  • research stays navigable over time

The long-term goal is to make scientific work more traceable, cumulative, and collaborative for both humans and AI.


Get Started


Join Us

OpenResearch may be interesting to you if you care about:

  • AI-assisted scientific discovery
  • human-AI research collaboration
  • structured scientific memory
  • interpretable research agents
  • long-horizon research workflows

Introduction

用户可借助该项目将研究组织为动态知识图谱,支持人机协作开展可追踪、可验证的 AI/ML 研究。核心功能包括声明-证据原子拆解、自动验证计划生成、实验结果结构化回写及图谱持续演化,实现研究过程的透明化与长期积累。【此简介由AI生成】

Customize your domain