Sec3 logo — Solana smart contract security firm
Back to Blog
Security

Introducing Owl LM (OwLLM v1): The First Open-Source Large Language Model for Web3

Sec3 Research Team

Today, we’re excited to open-source OwLLM v1.0 (colloquially called “Owl LM”), the first Open-source, web3-native Large Language Model. OwLLM has been trained on millions of transactions, has more than 100 million parameters, and is designed to be chain-agnostic.

Early LLM models, like ChatGPT, have been explored for web3 applications; however, their general-purpose nature may not be the best fit for a wide swath of addressable applications in web3 such as transaction analysis, threat detection, payment analytics, and more. To help devs harness the full potential of LLMs, we built an entirely new foundational model from scratch, one that was specifically designed to analyze and understand blockchain transactions.

In its current release, OwLLM has already proved itself capable of identifying MEV transactions with a 94% accuracy rate. As we continue to evolve and refine OwLLM—hopefully with the help of the community—we hope it will unlock new applications, such as conversational blockchain explorers, in-dApp MEV protection, (smarter) smart-contract security audits, malicious transaction monitoring, new regulatory compliance tools, emergent systematic/programmatic trading strategies, and more.

We're making OwLLM v1 open-source today to encourage developers to use and improve this model. To inspect the source code or start using OwLLM for your project, visit the repo here.

If you’re interested in exploring OwLLM for your project, would like to contribute novel data or insights, or have practical use cases for OwLLM, we'd love to hear from you: send an email to ai@sec3.dev. Join us in refining and expanding OwLLM!

Related Posts

Security

IDL Guesser

The Solana ecosystem thrives on innovation, but many Anchor-based programs do not publish up-to-date IDLs, which complicates the analysis of such programs and their transactions. To tackle this, we developed and open-sourced a prototype tool called IDL Guesser. This tool aims to automatically recover instruction definitions, required accounts (including signer/writable flags), and parameter information directly from closed-source Solana program binaries. This blog outlines the approach behind IDL Guesser and discusses potential areas for future improvement.

Read more
Security

All About Anchor Account Size

Smart contracts using Anchor require developers to allocate space for new accounts and specify the account size. Anchor provides guidelines for calculating the size based on the account structure, but many developers use std::mem::size_of instead, as they don't have to manually update the size when making changes to the account structure. Are they equivalent? In this blog post, we conduct a systematic comparison of the results produced by std::mem::size_of and the Anchor space reference.

Read more