Summary Scientify offers AI-powered research workflow automation, functioning as a continuous, evolving research partner. It targets researchers by automating literature review, hypothesis generation, and experimentation to achieve state-of-the-art results, enhancing scientific discovery efficiency.

How It Works It pioneers a "continuous metabolism" model, contrasting with batch AI tools. This involves constant literature ingestion, knowledge precipitation into a Git-managed Markdown knowledge base, hypothesis evolution, and proactive delivery of validated findings. A multi-agent system orchestrates iterative research pipelines (implementation, review, experiment), learning from failures. This approach demonstrably doubles hypothesis validation rates and significantly reduces token costs.

Quick Start & Requirements Requires Node.js >= 18, Python 3 with uv, and Git. Installation involves installing OpenClaw globally (pnpm add -g openclaw), onboarding (openclaw onboard), starting the gateway (openclaw gateway), and installing the Scientify plugin (openclaw plugins install scientify). Development requires cloning, installing dependencies, building, and linking. High-quality LLMs (e.g., Claude Opus 4.5+, GPT-5+) are recommended.

Highlighted Details

• Continuous Metabolism: Achieves 1.9x higher hypothesis validation rates and 92% lower token costs via ongoing literature processing.
• Autonomous Research: Capable of end-to-end research, autonomously discovering SOTA algorithms like KV2.
• Iterative Multi-Agent System: Employs specialized agents for iterative hypothesis testing, learning from failures.
• Auditable Markdown Knowledge Base: All research artifacts are stored as version-controlled Markdown files for full traceability.

Maintenance & Community Currently in internal beta, inviting researchers. Specific community channels or sponsorship details are not provided in the README.

Licensing & Compatibility Released under the MIT license, permitting broad usage, including commercial applications.

Limitations & Caveats Sub-agents have a 30-minute timeout, potentially limiting complex surveys. The OpenClaw sandbox lacks GPU passthrough. Research quality is highly LLM-dependent.