🌐 English | 中文文档
An AI with scientific curiosity that distributes experiments to humans — with immutable contribution tracking.
SciDAO is a three-layer architecture for AI-driven decentralized science:
┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐
│ Curiosity Engine │ → │ Task Marketplace │ → │Contribution Ledger│
│ AI generates │ │ Humans execute │ │ Immutable record │
│ hypotheses │ │ atomized tasks │ │ of every datapoint│
└─────────────────┘ └──────────────────┘ └─────────────────┘
- Curiosity Engine — AI explores a scientific domain and generates testable hypotheses
- Task Decomposer — Each hypothesis becomes atomized experiments anyone can run
- Contribution Ledger — Immutable, Git-like record of who did what
The AI has curiosity. You have hands. Let's do science together.
Three bottlenecks in current research:
| Bottleneck | SciDAO's Solution |
|---|---|
| Cognitive — literature, hypotheses, experimental design | AI does cognitive labor 24/7 |
| Execution — one lab can only run so many experiments | Atomized tasks → global executor network |
| Attribution — contributions buried in author lists | Immutable ledger → every datapoint traced to a person |
# 1. Clone
git clone https://github.com/SpencerRaw/sci-dao.git
cd sci-dao
# 2. Set API key
export DEEPSEEK_API_KEY=sk-... # or use OpenRouter:
export OPENROUTER_API_KEY=sk-...
# 3. Install dependencies
pip install openai
# 4. Run the curiosity engine
python scidao.py explore "carbon dots for cancer therapy"
# 5. Full pipeline: explore → decompose → marketplace
python scidao.py full "lipid nanoparticles for CRISPR delivery"python scidao.py explore <domain> # Generate hypotheses
python scidao.py full <domain> # Full pipeline
python scidao.py register <name> # Register as a contributor
python scidao.py submit <id> <task> <hypothesis> <result> # Submit results
python scidao.py leaderboard # Contribution rankings
python scidao.py lineage <hypothesis-id> # Trace hypothesis historyscidao/
├── scidao.py # CLI entry point
├── curiosity.py # Curiosity Engine: AI hypothesis generation
├── decomposer.py # Task Decomposer: hypothesis → atomized experiments
├── ledger.py # Contribution Ledger: immutable records
└── scidao/
└── llm.py # LLM client (DeepSeek / OpenRouter)
Dual-strategy hypothesis generation:
- Literature-grounded — explore a domain, generate falsifiable claims
- Self-critique → refinement — AI critiques its own hypotheses, proposes improvements
Deduplication by title similarity.
Each hypothesis becomes 3-5 atomized tasks with:
- Materials list, numbered protocol steps, expected results
- Time estimate + difficulty (BEGINNER / INTERMEDIATE / EXPERT)
- Inter-task dependencies
JSONL + SHA256 content hashing for immutable records:
register_contributor()— onboard a scientistrecord()— permanently record a contributionget_leaderboard()— rank contributors by weightget_hypothesis_lineage()— full provenance chainverify_integrity()— tamper-proof verification
No blockchain needed. Git-style integrity via content hashing.
Alpha — core pipeline functional. In development:
- Curiosity Engine (hypothesis generation + self-critique)
- Task Decomposer (atomized experiment protocols)
- Contribution Ledger (immutable JSONL + SHA256)
- Web interface for task marketplace
- Multi-agent debate ranking of hypotheses
- Experiment quality verification pipeline
- Lab execution platform integrations
- Contributions Never Forgotten — every datapoint traces to a person
- AI Has Curiosity — not a tool, a researcher
- Tasks Atomized — small enough for anyone to execute
- Trust Is the Moat — immutable contribution records
- Google's AI Co-Scientist — AI hypothesis generation
- AI Scientist concept by Garrick Hileman
- Git's content-addressable integrity model
- The open science movement
MIT — see LICENSE.
"Nature is the judge. You are the jury. The AI is the prosecutor."
Issues and PRs welcome. Before submitting:
pip install -e "."
pytestSee PLAN.md for roadmap and design decisions.