NOVA for Finance: : Secure DCA Memory with Self-Learning Loops and Confidential Portfolio Analysis

Users securely store their Dollar-Cost Averaging (DCA) trading history in a personal, decentralized vault powered by NOVA. By connecting their NEAR wallet, they create a group (e.g., "dca-history"), upload encrypted JSON memos of weekly trades, and dynamically retrieve/update the history. This data is made available to Shade Agents (TEE-based) for confidential portfolio analysis, risk modeling, and self-learning loops (e.g., fine-tuning AI models on historical trades for optimized asset allocation suggestions—without ever exposing plaintext data).

Use Case

Personal Crypto Portfolio Management (e.g., Weekly $100 DCA)

In a self-managed DCA strategy, users upload a JSON array of trade memos weekly to their personal NOVA vault (e.g., "dca-history.{user_account_id}" for uniqueness). Each entry includes date, operation, provider, portfolio snapshot, and rationale.

When performing a weekly update (e.g., on December 31, 2025):

  • The app retrieves the latest encrypted data from IPFS via NOVA's SDK.

  • Decrypts it locally using the group key from Shade Agents (TEEs).

  • Appends a new entry (e.g., {"date": "31-12-2025", "operation": "Swapped 50-worth-usd of NEAR for ETH.", "provider": "https://dex.example.com", "portfolio": "4.5 NEAR, 1,800 stNEAR, 0.8 SOL, 0.001 BTC, 0.05 ETH", "rationale": "Diversifying into ETH ahead of expected market recovery."}).

  • Re-uploads the updated JSON as a new encrypted version to IPFS, logging metadata on NEAR for versioning.

  • For analysis: The app invokes a Shade Agent to process the full history confidentially—e.g., compute risk metrics, simulate scenarios, or fine-tune a lightweight model for future recommendations—outputting verifiable attestations (e.g., "Portfolio risk: Low; Suggested next DCA: 60% NEAR, 40% SOL") without data leaks.

This enables automated, privacy-preserving portfolio management, with self-learning loops where agents iteratively improve based on user history, all under user sovereignty.

How NOVA matches the needs of a personal trading assistant

  • User Sovereignty & Privacy: Personal groups ensure only the user (and temporarily granted agents) access data. Encryption is client-side, with keys in TEEs (Shade Agents) for zero-trust analysis.

  • Versioned History with One-Way Updates: Retrieve and append to JSON without re-encrypting old data; new uploads create immutable versions on IPFS, tracked on NEAR.

  • Confidential Analysis via TEEs: Shade Agents process decrypted data in secure enclaves for portfolio metrics, risk assessment, or AI fine-tuning—e.g., self-learning loops that adapt models to user patterns without exposing history.

  • Verifiable & Auditable: All uploads/logs on NEAR blockchain for tamper-proof auditing; agents provide attestations for transparency.

  • Scalability for Weekly Ops: Low-cost NEAR transactions (~0.01 NEAR/tx) suit frequent updates; IPFS ensures efficient storage/retrieval.

  • AI Agent Integration: Use NOVA's MCP (Model Context Protocol) to invoke agents for natural language queries like "Analyze my DCA history for Q4 2025" or self-learning (e.g., "Fine-tune allocation model on my trades").

  • Extensible for DeFi Automation: Integrate with NEAR wallets for seamless wallet connects; extend to trigger DCA ops via intents while logging securely.

  • Self-Learning Loops: Agents can iteratively refine models (e.g., via federated learning proxies) on encrypted history, improving suggestions over time without central servers.

This turns NOVA into a backbone for privacy-first personal finance apps, empowering users with confidential AI-driven insights.

Running the Demos

  • Set .env file

  • Rust: cargo run --bin dca-history.

  • JS: ts-node dca-history.ts.

Expected output:

Last updated