NeuriCo: Hypothesis In, Research Paper Out

Key Takeaways

• NeuriCo (Neural Co-Scientist, inspired by Enrico Fermi) is an autonomous research framework from the University of Chicago Human+AI Lab
• Minimal input, maximum output: provide a title, domain, and hypothesis — agents handle literature review, experiment design, code execution, analysis, and LaTeX paper writing
• Multi-provider: works with Claude Code, Codex CLI, and Gemini CLI — no vendor lock-in
• IdeaHub integration: community platform where anyone can submit research ideas for agents to explore — fetch an idea and run it with one command
• Docker-ready: pre-built image with GPU support, CLI tools, LaTeX, and paper-finder pre-installed
• Auto GitHub integration: creates repos and pushes results automatically
• Repository: ChicagoHAI/NeuriCo

What Is NeuriCo?

NeuriCo is an autonomous research framework that takes structured research ideas and orchestrates AI agents to design, execute, analyze, and document experiments across diverse domains.

Core attributes:

• Category: Autonomous end-to-end research framework
• Origin: University of Chicago, Human+AI Lab
• Name: Neural Co-Scientist (inspired by Enrico Fermi)
• License: Apache 2.0
• Stars: 63+
• Providers: Claude Code, Codex CLI, Gemini CLI

What problem does it solve?

• Existing research agents require complex setup and manual orchestration
• NeuriCo reduces the input to three fields: title, domain, hypothesis
• Everything else — literature search, baseline identification, coding, analysis, writing — is handled by the agent

How It Works

# Install (one-liner: clones repo, pulls Docker, walks through setup)
curl -fsSL https://raw.githubusercontent.com/ChicagoHAI/neurico/main/install.sh | bash

# Fetch an idea from IdeaHub and run it
cd NeuriCo
./neurico fetch <ideahub_url> --submit --run --provider claude --write-paper --full-permissions

That's it. The agent fetches the idea, creates a GitHub repo, runs experiments, writes a paper, and pushes everything.

The Pipeline

Research idea (title + domain + hypothesis)
    ↓
Literature review (Semantic Scholar + paper-finder)
    ↓
Dataset search + baseline identification
    ↓
Experiment design
    ↓
Code generation + execution (Docker sandbox)
    ↓
Result analysis + figure generation
    ↓
LaTeX paper writing
    ↓
GitHub repo creation + push

IdeaHub: The Community Research Platform

What makes NeuriCo unique is IdeaHub — a community platform at hypogenic.ai/ideahub where anyone can submit research ideas for agents to explore.

How it works:

• Browse community-submitted research ideas
• Pick one (or submit your own)
• Run it with NeuriCo: ./neurico fetch <url> --run
• The agent does the rest

Why this matters:

• Democratizes research — you don't need to be an expert to propose an idea
• Creates a shared pool of research questions
• Enables reproducibility — anyone can re-run the same idea

Multi-Provider: No Vendor Lock-in

All providers use OAuth login — no API keys needed for the agent itself. Optional API keys (OpenAI, Semantic Scholar, Cohere, HuggingFace, W&B) enhance functionality.

Setup Tiers

NeuriCo vs Other Research Agents

NeuriCo's edge: IdeaHub integration + Docker isolation + multi-provider flexibility. Less battle-tested than AutoResearchClaw (9.4K stars) but more modern architecture and the community ideas platform is genuinely novel.

Docker Setup

Docker is the recommended deployment — isolated environment with GPU support and all tools pre-installed:

# Pull pre-built image (~2 min)
docker pull ghcr.io/chicagohai/neurico:latest

# Configure
./neurico config

# Login to your AI CLI (one-time)
claude  # or: codex, gemini

# Run
./neurico fetch <ideahub_url> --run --provider claude --write-paper

The Docker image includes: Python venvs, CLI tools (Claude/Codex/Gemini), LaTeX, paper-finder, and all dependencies.

FAQ

Q1: How is NeuriCo different from just asking ChatGPT to write a paper?

NeuriCo actually executes code, runs experiments, generates real figures from real data, and produces a complete LaTeX manuscript. ChatGPT generates text but doesn't execute anything.

Q2: What domains does it support?

Domain-agnostic — ML, data science, AI, systems, theory, biology, physics, and more. The agent adapts to whatever domain the research idea specifies.

Q3: Can I use my own research idea instead of IdeaHub?

Yes. Create a YAML file with title, domain, and hypothesis, then run ./neurico run --idea my_idea.yaml --provider claude.

Q4: Does the Docker container need a GPU?

GPU is optional but recommended for ML experiments. The container works on CPU for non-ML research tasks.

Q5: How does paper-finder work?

Paper-finder uses Semantic Scholar API to search for relevant papers, with optional Cohere reranking for improved relevance (~7% improvement). It automatically generates a literature review section.

Q6: Is the generated paper publishable?

NeuriCo produces well-structured drafts with real experimental results. Human review and refinement are still needed for publication quality, but it handles 70-80% of the mechanical work.

Summary

NeuriCo represents the next generation of autonomous research tools — minimal input (just a hypothesis), flexible execution (three AI providers), community-driven ideas (IdeaHub), and production-grade isolation (Docker + GPU). While younger than alternatives like AutoResearchClaw, its architecture is more modern and the IdeaHub platform introduces a genuinely new paradigm: democratized research ideas that AI agents can explore. For researchers who want to rapidly prototype research directions, NeuriCo is the fastest path from idea to paper.