Vaultra

Vaultra is an autonomous onchain treasury agent that transforms passive crypto holdings into an actively managed, intelligent portfolio. Instead of manually monitoring balances, reallocating assets, or executing trades, users can delegate treasury decisions to Vaultra, which continuously analyzes portfolio composition, detects inefficiencies such as idle stablecoins, and proposes optimized strategies based on user-defined risk preferences.

Through a natural language interface, users can interact with Vaultra like a CFO issuing instructions such as “optimize treasury” or “reduce volatility.” The system responds with structured plans, including reasoning, expected outcomes, and actionable steps. Once approved, Vaultra executes these strategies onchain, ensuring transparency and real-time updates to the portfolio.

Vaultra also maintains a persistent memory layer, allowing it to learn user behavior and refine future decisions. By combining AI-driven reasoning, onchain execution, and modular agent architecture, Vaultra enables individuals, DAOs, and teams to manage capital more efficiently, proactively, and autonomously.

Vaultra is built as a full-stack application combining a modern frontend, an AI-driven backend, and a lightweight onchain execution layer. The frontend is developed using Next.js, TypeScript, Tailwind CSS, and shadcn/ui, delivering a high-performance and polished user experience with real-time portfolio visualization and a ChatGPT-style interaction interface.

The backend is implemented using Node.js with Express, acting as the orchestration layer for AI reasoning, portfolio analysis, and transaction routing. We integrated large language models to power Vaultra’s decision-making, where structured prompts transform raw portfolio data into actionable treasury strategies. These strategies are then parsed into executable steps.

For onchain interaction, Vaultra integrates with Uniswap to generate swap routes and transaction calldata, enabling real asset rebalancing. Execution reliability is handled via a dedicated executor layer (and optionally KeeperHub), ensuring transactions are broadcasted, retried, and logged properly. A minimal smart contract is deployed to simulate and log treasury actions, improving transparency and auditability.

We also incorporated ENS to provide human-readable identity for agents, and designed the system to support persistent memory storage (e.g., via 0G) for storing preferences, history, and agent state. One notable hack was abstracting complex DeFi flows into simple AI-generated plans, allowing users to trigger multi-step financial actions through a single natural language command, dramatically simplifying the UX while preserving real onchain execution.