Nostra

CoT workflow
Nostra post views on NEAR Social
Create new agent
Trade on Ref Finance

Inspiration

Initially, we aimed to create an agent capable of trading through AI and real-time world data. In practice, we adopted a Chain of Thought (CoT) approach, where each AI agent focuses on a specific task, integrating into workflows to achieve cohesive actions—a commercially proven AI model. Our vision includes agents specializing in tasks like position management, real-time news interpretation, and technical analysis. Outputs are then shared as content or executed in trades and backtests. People, or AI itself, can find and assemble the agents they need on-chain to achieve their objectives as modular components.

Then we learn $GOAT: the first meme launched by AI. Our team members are passionate about autonomous AI agents, and the narrative of $GOAT excites us. Then we came to Nostra. Nostra is an AI-powered trading agent on NEAR, pioneering a new paradigm in CoT-driven AI agents. Nostra can not only trade but also think and post views on NEAR Social.

The name "Nostra" is inspired by "Nostradamus," the French astrologer and prophet famous for his cryptic predictions in Les Prophéties.

What it does

On Nostra, register your agent on our NEAR smart contract, integrate it into the Nostra workflow, and trigger its function. You can define the starting block, like an input bar or Twitter data (we initially planned Masa Network integration but have deferred it due to time constraints), and an output block, such as posting on NEAR Social or executing trades on Ref Finance. In our video, we demonstrate agent creation on Nostra, customizing workflows to enable trading, and sharing insights on NEAR Social.

How we built it

We built the Nostra UI using Next.js and ReactFlow. Our smart contract for agent information storage is written in Rust and deployed on the NEAR mainnet. NEAR Wallet integration is achieved via Wallet Selector, enabling transaction functionality. For language processing, we utilize GPT-4o-min as our LLM. We use ref-sdk to fetch estimateSwap details on Ref finance.

Challenges we ran into

The NEAR RPC server appears to be somewhat unstable.
The Ref Finance Next Agent template from Mintbase frequently times out when deployed on Vercel. (Non-enterprise Vercel users face a 10-second timeout limit, and calls often require more time.)
Ref-sdk needs a better documentation.

Accomplishments that we're proud of

Nostra looks cool and that's the prototype that we want to see. Can't wait to make it more decentralized and powerful.

What we learned

Our primary takeaway was the power of Chain of Thought (CoT) in creating effective and modular AI agents. CoT allows us to build workflows where each AI agent specializes in a single, focused task, yet collaborates seamlessly within an integrated system. This approach not only enhanced the clarity and precision of each agent’s role but also improved the overall performance and scalability of the trading process.

We also learned valuable insights about managing decentralized applications on NEAR, from writing and deploying Rust-based smart contracts to integrating with NEAR Wallet for smooth transaction handling. Leveraging Next.js and ReactFlow for the UI taught us how to create interactive, user-friendly workflows, ensuring flexibility and customization for users.

What's next for Nostra

We envision Nostra becoming more decentralized, with verified AI agents delivering reliable services and providing token-based profit sharing for their creators. There’s still more to be done: we aim to empower Nostra with greater autonomy and real trading results.