Your research deserves better tools.
Open Research is an AI research director that lives in your terminal. It searches papers, critiques methodology, designs experiments, and drafts publications — from literature review to revision letter.
Built for researchers who think in hypotheses, not functions.
npm install -g open-research
The problem
Cursor and Claude Code are for engineers. You need an agent that understands research.
General-purpose AI assistants don't understand evidence hierarchies, threats to validity, or the pain of a major revision. Research demands different tools.
Literature review takes months, not hours
You are reading 200+ papers for a systematic review. You need to track inclusion criteria, assess methodology quality, extract data, resolve conflicts between studies. ChatGPT gives you a bulleted summary with hallucinated citations.
Evidence synthesis requires judgment, not generation
Three meta-analyses reach different conclusions about the same intervention. You need to understand why -- different inclusion criteria? Different statistical methods? Publication bias? Generic AI tools cannot reason about evidence hierarchies.
Responding to peer review is soul-crushing
Reviewer 2 raises 14 concerns across your methodology, statistics, and framing. You need point-by-point responses backed by your actual data, new analyses, and literature. This takes weeks of context-switching between tools.
Full lifecycle coverage
From first idea to camera-ready paper.
Every phase of the research process has a dedicated skill. One agent, eight phases, zero context switching.
/novelty-checkerValidate novelty/source-scoutFind every paper/devils-advocateStress-test claims/experiment-designerDesign with rigor/data-analystStats & plots/draft-paperLaTeX drafting/methodology-criticSelf-audit/reviewer-responseAddress reviews
Capabilities
14 tools. 11 skills. Three agent modes.
Full filesystem and shell access, federated paper search, two-tier memory, live LaTeX preview, and research-specific skills that encode real methodology.
Federated Paper Search
Search arXiv, Semantic Scholar, and OpenAlex simultaneously. Cross-reference results and rank by relevance, citation count, and recency.
11 Research Skills
From novelty checking to reviewer response generation. Each skill encodes real research methodology grounded in how PIs actually work.
Sub-Agent Delegation
Spawn lightweight explore agents with their own context windows. They navigate literature and datasets independently, then report back.
Smart Context Management
Automatic compaction at 90% context usage. Preserves critical research findings and citations while freeing space for deeper exploration.
Local-First Workspace
Everything in your filesystem: sources/, notes/, papers/, experiments/. Works with git. No cloud dependency. Your data never leaves your machine.
3 Agent Modes
Manual review for careful work. Auto-approve for trusted operations. Auto-research with a Research Charter for fully autonomous investigation.
Knowledge Ontology
An automatic knowledge graph that structures findings into connected, evidence-traced notes. Your research compounds — every session builds on the last.
Research Ontology
Your knowledge compounds. Every session builds on the last.
Open Research automatically builds a structured knowledge graph as you work. Every paper read, claim made, finding extracted, and method discovered gets captured as connected, evidence-traced notes — not flat files.
How it works
After each turn
A background ontology manager extracts knowledge from the conversation and structures it into typed notes and edges.
Before each turn
A relevance agent selects notes related to your current question and injects them as context — you always have what you need.
During a turn
The agent queries the ontology for evidence, contradictions, and connections across everything you've ever researched.
What it captures
sourcePapers, URLs, datasets
findingResults from sources
claimArguments & assertions
questionOpen gaps & unknowns
methodTechniques & approaches
insightCross-finding synthesis
Connections
Each edge has a strength (strong / moderate / weak) and a context explaining why the connection exists.
Built-in skills
Research methodology, encoded.
Each skill encapsulates expert research practices. These are not prompt wrappers. They are structured workflows that a principal investigator would recognize.
/novelty-checkerEvaluate if your research idea is truly novel against the existing literature. Searches across databases and identifies prior work you might have missed.
"Has anyone studied the effect of sleep deprivation on LLM evaluation quality?"
/source-scoutFind and rank the most relevant papers you are missing for any research question. Builds citation graphs to surface influential but overlooked work.
"Find seminal papers on causal inference in observational epidemiology."
/paper-explainerDeep-read complex papers. Generates structured explanations, comparison tables, and identifies methodological choices you should be aware of.
"Explain the key differences between DPO, RLHF, and RLAIF training approaches."
Getting started
Three commands to grounded research.
Initialize your workspace
Create a structured research workspace with directories for sources, notes, papers, and experiments. Everything stays local.
> /init-- workspace/ with sources/, notes/, papers/, experiments/
Ask your research question
The agent searches across arXiv, Semantic Scholar, and OpenAlex. It reads papers, assesses methodology, and synthesizes findings autonomously.
~ Searching arXiv, Semantic Scholar... found 64 papers
~ Reading top 12 by citation count and methodological rigor...
Get citable, grounded results
Analysis, structured notes, and BibTeX entries are saved directly to your workspace. Everything is traceable to real papers. No hallucinated citations.
notes/rlhf-failure-modes.md-- structured analysis with evidence grades
sources/rlhf-papers.bib-- 64 verified BibTeX entries
notes/rlhf-taxonomy.md-- comparison table across 12 key papers
Install
Start researching in 10 seconds.
One global install. Works anywhere Node.js runs. Then launch, authenticate, and go.
npm install -g open-researchRequires Node.js 20+
Quick start
Open source. Local-first. No data leaves your machine.
MIT licensed. No telemetry. No cloud dependency. Every paper search, every analysis, every draft stays in your local workspace.
Fully Open Source
MIT licensed. Read every line of code. Fork it, extend it, make it yours.
Local-First
Everything runs on your machine. Your research stays your research.
Zero Telemetry
No data collection. No tracking. No analytics. Just research.