Scientify
Open-Source, Free Research Agent

Let OpenClaw Do Research
for You, 24/7

Stop spending weeks reading papers and debugging code. Type a research topic, and 6 AI agents relay from literature survey all the way to experiment validation.

Install with one command

openclaw plugins install scientify
Get StartedView on GitHub

What Scientify Can Do for You

Powered by OpenClaw, one prompt triggers a complete research workflow

End-to-End Research Execution
“Research scaling laws for classical ML classifiers on Fashion-MNIST”

The orchestrator chains 6 phases automatically: literature search → deep analysis → plan design → code implementation → review iterations → full experiments. Each phase runs as an independent sub-agent.

Search Journals, Preprints & GitHub
“Get the latest papers on vision-language models for medical imaging”

Search arXiv, OpenAlex, and GitHub in parallel. Download .tex/.pdf full-text files, auto-cluster by sub-topic, and output a structured survey report.

Generate & Evaluate Research Ideas
“Explore recent advances in protein folding and generate innovative research ideas”

Generate 5 research ideas grounded in real papers, scored on novelty, feasibility, and impact. The top idea is auto-expanded into a full proposal with detailed methodology.

Remember Your Research, Struggles & Progress
“My topic was recently covered in new ICLR research — look into it and let's discuss.”

Each topic gets its own persistent workspace — papers, analyses, and code all saved. Supports project switching, progress queries, and resuming from breakpoints with full cross-session context.

6-Phase Research Pipeline

Each phase runs as an independent AI agent. The orchestrator verifies outputs between steps and passes context forward.

Phase 1

Literature Survey

Search arXiv + OpenAlex, filter, download .tex sources, cluster by direction

survey/report.md
Phase 2

Deep Analysis

Extract formulas, map methods to code, build cross-comparison table

survey_res.md
Phase 3

Implementation Plan

Design 4-part plan — Dataset / Model / Training / Testing

plan_res.md
Phase 4

Code Implementation

Write ML code in uv-isolated venv, validate with 2-epoch run

project/run.py
Phase 5

Automated Review

Review code, fix issues, rerun, re-review (up to 3 rounds)

iterations/judge_v*.md
Phase 6

Full Experiment

Complete training + ablation studies with final analysis

experiment_res.md

Power Moves

Combine with MCP servers, browser automation, and multi-session concurrency to unlock more

Don't Want to Miss New Papers? Read & Think for You

“Based on my topic, automatically search for new papers daily, download and read them, figure out which ones are inspiring, and send conclusions to Slack/Feishu”

Pair with scheduled triggers and MCP integrations (Slack/Feishu/Email) to auto-track your field daily. Never miss an important paper again.

Scheduled trigger
arxiv + openalex search
LLM filtering + summary
Push to Slack/Feishu

Can't Download Paywalled Papers? Use Campus Access

“Download those papers you found to my workspace”

Playwright MCP opens a browser, authenticates through your institutional proxy, and batch-downloads paywalled papers. No more clicking one by one.

Scientify provides URLs
Playwright opens browser
Institutional auth
Download PDF

Multi-Branch Exploration? Run in Parallel

“Help me analyze which of 3 methods has the most potential: LoRA fine-tuning, MoE architectures, KV-Cache optimization”

Launch multiple research pipelines concurrently with one prompt. Each runs in its own directory without interference, with a final cross-topic analysis.

Main orchestrator
Sub-session per topic
Independent pipelines
Cross-topic analysis

Can't Follow the Math? Section-by-Section Walkthrough

“Walk me through 'Attention Is All You Need' section by section, explain every formula”

Read papers page by page. AI explains every formula derivation, compares approaches from related work, and finds open-source implementations for you.

Page-by-page browsing
Formula derivation
Compare methods
Find implementations

Want to Reproduce Results? Auto-Run

“Reproduce the results from Table 2 of this paper”

AI deep-reads the paper, extracts experiment design, auto-generates code and runs it, then puts your results side-by-side with the original numbers.

Deep read paper
Extract experiment design
Write & validate code
Run experiment
Compare results

Skills, Tools & Commands

Everything included in the plugin

research-pipeline

Orchestrator. Spawns sub-agents for each phase, verifies outputs between steps.

literature-survey

Search arXiv, filter, download .tex sources, cluster, generate survey report.

research-survey

Deep analysis: extract formulas, map to code, produce method comparison table.

research-plan

Create 4-part implementation plan (Dataset/Model/Training/Testing).

research-implement

Implement ML code from plan, run 2-epoch validation with uv venv isolation.

research-review

Review implementation. Iterates fix, rerun, review up to 3 times.

research-experiment

Full training + ablation experiments. Requires review PASS.

idea-generation

Generate 5 innovative research ideas, select and enhance the best one.

write-review-paper

Draft a review/survey paper from project research outputs.

Get Started in 3 Steps

Up and running in under 5 minutes

1

Install OpenClaw

terminal
pnpm add -g openclaw
openclaw onboard
openclaw gateway
2

Install Scientify

terminal
openclaw plugins install scientify
3

Start Researching

terminal
# Open WebUI at http://127.0.0.1:18789
# Type your research prompt:
Research scaling laws for classical ML classifiers on Fashion-MNIST