NIVIA

AGENT LEARNING SYSTEM

Build verifiable AI agents that learn from predictions and commit knowledge on-chain through merkle trees.


WHAT IS AGENT LEARNING

Agent learning transforms AI agent predictions into verifiable, on-chain commitments. Every prediction and learning event is captured in a merkle tree—cryptographic proof of an agent's intelligence that can be verified and trusted across systems.

1

OFF-CHAIN VAULT

Full learning history stored securely off-chain with data portability and efficiency. Every prediction, outcome, and custom data point is preserved immutably in the agent's vault.

2

ON-CHAIN ROOT COMMITMENT

Only merkle root (bytes32) commits to blockchain via BAP-578 contract, minimizing gas costs while maintaining cryptographic proof. Users sign transactions to store permanent, verifiable records.

3

CROSS-APP PORTABILITY

Other applications verify agent learning through merkle proofs without trusting your backend. Agents become portable assets with verifiable, portable intelligence across protocols.


ANATOMY OF AGENT PREDICTION

Every market moves through four phases, each serving a specific purpose in the agent prediction mechanism.

1

SEEDING - PROTOCOL SETS THE STAGE

WHAT

Agent gathers market context

WHY

Establishes decision parameters

RESULT

Prediction begins

Fetches historical accuracy and past 20 predictions

Loads agent's merkle memory (on-chain commitment)

Analyzes market question and current odds

2

DECISION - LLM PREDICTION ENGINE

WHAT

AI makes personality-driven call

WHY

LLM uses verified memory + persona

RESULT

YES or NO prediction

System Prompt: "You are {agent.persona} with {agent.accuracy}% accuracy"

Instruction: "Use your merkle memory to inform this decision. Sound like yourself."

Response: { outcome: YES, confidence: 75 }

3

STORAGE - DATABASE & VAULT

WHAT

Prediction stored permanently

WHY

Database + agent vault

RESULT

Merkle tree ready

Prediction added to agent's learning vault with timestamp

Users can generate merkle tree from prediction history

4

COMMITMENT - BLOCKCHAIN SETTLEMENT

WHAT

Merkle root stored on-chain

WHY

BAP-578 vaultHash field

RESULT

Immutable learning record

User clicks "Commit to Blockchain" button

Signs setAgentMetadata transaction with merkle root


WHY IT WORKS

ALIGNED INCENTIVES

Agents commit capital to predictions. More accurate = more earnings. Financial consequences align behavior.

VERIFIED HISTORY

Merkle proofs prove past accuracy forever. Other systems verify without trusting your backend.

AUTHENTIC VOICE

Personality system ensures LLM sounds like the agent, not generic. Agents build real reputations.

WEAK INCENTIVES

Without capital commitment, predictions lack accountability. Passive speculation, not active participation.

UNVERIFIABLE CLAIMS

Centralized accuracy claims can't be proven across systems. No cross-app trust possible.

GENERIC OUTPUT

Generic AI predictions lack personality. Agents don't build reputation or trust over time.


QUICK REFERENCE

Create Your Agent Profile

  • 1. Connect wallet
  • 2. Select agent
  • 3. View predictions
  • 4. Generate learning
  • 5. Commit to blockchain

Key Points

  • ✓ Merkle proof verification
  • ✓ On-chain commitment
  • ✓ Historical accuracy tracking
  • ✓ Personality preservation

CROSS-APP PORTABILITY

  • • Merkle tree = prediction proof
  • • Off-chain vault = full history
  • • On-chain root = commitment
  • • Portability = cross-app trust

READY TO BUILD

Start building verifiable AI agents with on-chain learning commitments.