Welcome

About

Aubrai is the world’s first decentralized scientific agent with knowledge stemming from thousands of private lab notes, internal chats and unpublished insights from the lab of Dr. Aubrey de Grey and the collective intelligence of the global longevity community.

Co-developed by VitaDAO and BIO, the first BioAgent is designed to fight the greatest killer of all time: Aging. As an onchain AI co-scientitst, Aubrai can generate and validate hypotheses, design wet-lab experiments and encrypt data when asked, enriching research outputs while protecting trade secrets.

At the heart of Aubrai's mission lies the Robust Mouse Rejuvenation (RMR2) project – Aubrey's ambitious study to double the remaining lifespan of middle‑aged mice. If successful, it could be aging's "AlphaFold moment": a proof that multi‑target rejuvenation works and is worth scaling.

Architecture Overview

Aubrai is built on the BioAgents framework powered by ElizaOS and purpose-built for longevity science. It layers domain-specific prompts / tools and curated longevity literature on top of the BioAgents framework.

Because BioAgents is under active development, Aubrai continuously inherits upstream improvements, while adding domain-specific upgrades. This approach ensures Aubrai's intelligence capabilities compound and grow with every release.

Aubrai extends core components from the BioAgents framework, optimized for the domain of longevity research. Its key intelligence building blocks:

• Internal knowledge

• Longevity knowledge graph

• OpenScholar fine-tuned on longevity literature

Internal Knowledge

Aubrai has been trained on a private collection of documents from the LEV Foundation. These include but are not limited to

• Books

• Editorials

• Emails

• Notes

• Published and unpublished papers

• Results from the RMR study

The result is an agent which functionally personifies Aubrey de Grey as a digital clone.

Knowledge Graph

The knowledge graph acts a semantic representation of longevity research, powered by the Longevist, a VitaDAO curated library spanning over 4000 papers. While reasoning, the agent considers if it needs more information to refine its response. In these cases, it queries the knowledge graph to collect further information, enriching the final output with deeper scientific insights.

OpenScholar

Aubrai's scientific outputs are further refined using a fine-tuned version of OpenScholar. The model weights used in Aubrai for the OpenScholar retriever and reranker models can be accessed through the Bio Protocol Huggingface page.