How AI Powers Bundie

Bundie uses AI for intelligence, not automation. Our AI is your personal DeFi research assistant that analyzes opportunities and makes recommendations - but you stay in complete control of every decision.

What AI Does

1. Analyzes Your Wallet History

The AI scans your on-chain activity across supported EVM chains to understand your DeFi behavior:

Transaction patterns: What protocols you've used across DeFi
Risk profile: Types of strategies you prefer (lending, LP, staking)
Position sizing: Average amounts you typically invest
Hold duration: How long you typically keep positions
Chain preferences: Which networks you're most active on

Analysis time: ~3 minutes for a comprehensive profile. Results are cached for 7 days, so repeat recommendations are near-instant.

New wallet? If you have no DeFi history, the AI builds a curated starter bundle and learns your preferences as you use Bundie.

2. Builds Your Best Bundle

Based on your wallet activity, the AI assembles the best-performing strategies into a single optimized bundle using a multi-agent debate:

A Bull agent argues for the highest-APY opportunities
A Bear agent argues for the safest, highest-risk-score venues
A Moderator agent synthesizes both into a risk-adjusted bundle

A fairness step then spreads exposure so a bundle isn't dominated by a single protocol. The default bundle holds 3 strategies, and new users get a curated starter bundle designed for a smooth first experience.

There are no separate "risk tiers" to choose from. The AI does the work and gives you its best recommendation. You can always browse all strategies and build your own bundle if you prefer.

3. Scores and Recommends Strategies

The AI continuously evaluates yield strategies across every integrated protocol:

What it evaluates:

Current APY and historical trends
Protocol maturity and track record
Total Value Locked (TVL) and liquidity depth
Smart contract security and audit status
Correlation between different strategies
Gas costs vs potential returns

Scoring system:

Each strategy gets a 0-100 risk score (higher = safer), built from four components: security, liquidity, maturity, and centralization
Recommendations are filtered to your profile band — Conservative (≥85), Moderate (≥75), or Aggressive (≥70)
An AI risk-enrichment pass can adjust component scores after aggregation; results are cached and refreshed regularly

What AI Doesn't Do

No Automatic Trading

You approve every transaction
AI cannot move your funds without your explicit consent
All allocations and withdrawals are user-initiated

No Rebalancing Without Your Say-So

The AI can compute rebalance targets and monitor rule-based triggers, but it never moves funds on its own
Rebalances and rule automations only run when you opt in and approve them
You decide whether to act on any recommendation

No Custody

Your funds stay in your self-custodial account
AI has no access to your private keys
You retain complete control at all times

No Guaranteed Returns

AI recommendations are based on historical data and current conditions
Past performance doesn't guarantee future results
DeFi yields are variable and can change rapidly

AI Recommendations vs Your Control

Think of Bundie AI as a research assistant, not a portfolio manager:

AI's Role	Your Role
Suggests strategies	Decides which to use
Analyzes risk	Sets risk tolerance
Monitors performance	Initiates withdrawals
Recommends bundles	Allocates capital
Provides insights	Makes final decisions

AI Transparency

What Data We Use

On-chain data: Public transaction history from your wallet
Protocol data: TVL, APY, audit status from public sources
Market data: Price feeds, liquidity depth, volatility metrics

Analysis and recommendations are produced by large language models (Google Gemini 2.5 for analysis and bundle debate; Claude for the options engine), always grounded in the on-chain and protocol data above.

What We Don't Use

✗ Private information or personal data
✗ Off-chain activity or browsing history
✗ Email contents or social media
✗ Other users' strategies (no copy-trading)

How We Protect Your Privacy

All analysis happens on-chain or using public data
No personal information required for AI analysis
Wallet addresses are pseudonymous
No data sold to third parties

Improving AI Recommendations

The AI learns and improves over time:

Feedback Loop

Your actual investment choices train the model
If you consistently ignore certain recommendations, AI adapts
Successful strategies are weighted higher for similar users

Community Learning

AI improves as more users join the platform
Better understanding of risk profiles across user segments
More accurate APY predictions based on historical data

Next Steps

Understanding Your Vault Journey - Complete guide from setup to withdrawal
Want to build on Bundie's AI? See the Developer Docs for the MCP server, Agent Skills, and API reference.

Ready to get personalized recommendations? Launch Bundie App →