Giza ZKML Agent Visualization

Overview

Giza is a protocol for deploying verifiable AI agents on-chain using Zero-Knowledge Machine Learning (ZKML). This skill covers how Giza agents work, proof verification, and how to interpret the visualization dashboards in SperaxOS.

Key Concepts

Zero-Knowledge Machine Learning (ZKML)

ZKML allows AI model inferences to be verified on-chain without revealing the model weights or input data. This creates trustless AI — anyone can verify that a model produced a specific output from a specific input, without needing to trust the model operator.

Giza Agents

On-chain AI agents deployed via the Giza protocol. Each agent:

• Runs a specific ML model (e.g., price prediction, risk scoring)
• Generates ZKML proofs for each inference
• Can be deployed on multiple chains (Starknet, Ethereum, Arbitrum, etc.)
• Has a verifiable track record of accuracy and performance

Proof Systems

Giza supports multiple proof backends:

• Cairo — Native to Starknet, fastest proving time for StarkNet deployments
• Noir — Aztec Labs' DSL for ZK circuits, good for Ethereum L1
• RISC Zero — General-purpose zkVM, supports any computation

SperaxOS Visualization Tool

Agent Overview Dashboard

Shows all deployed Giza agents with:

• Status indicators — Green (active), Yellow (pending), Red (inactive)
• Summary stats — Total active agents, inference count, proof count
• Agent rows — Name, chain, inference count, proof count per agent

Proof History View

Visualizes ZKML proof verification pipeline:

• Verification rate bar — Color-coded segments showing verified/pending/failed ratios
• Proof rows — Individual proofs with status badge, proof type (Cairo/Noir/RISC0), chain, duration, and timestamp
• Use this to monitor proof verification health and identify failures

Model Performance Dashboard

Detailed metrics for a specific AI model:

• Accuracy gauge — Green >95%, Yellow 85-95%, Red <85%
• Inference latency — Average time per inference
• Proof generation time — Average time to generate ZKML proof
• Inference volume sparkline — 30-day trend of inference activity
• Chain deployment tags — Which chains the model is deployed on

Protocol Analytics

Protocol-wide dashboard with:

• Hero stats — Total agents, active agents, total proofs, 24h proof count
• Chain distribution bars — Horizontal bars showing agent and proof distribution per chain
• Dual trend chart — Overlapping proof volume (solid purple) and agent count (dashed green) trends

Common Use Cases

1. "Show me Giza AI agents" → Agent Overview
2. "What's the proof verification rate?" → Proof History
3. "How is model X performing?" → Model Performance (requires modelId)
4. "Give me Giza protocol stats" → Protocol Analytics
5. "Which chains have the most Giza agents?" → Protocol Analytics chain breakdown

Technical Details

API Integration

The tool connects to api.gizatech.xyz/api/v1 with:

• Automatic retry with exponential backoff (2 retries)
• 12-second timeout per request
• Graceful fallback to curated demo data when API is unavailable

Data Freshness

• Agent and proof data is fetched in real-time from Giza's API
• Demo data is deterministically generated for consistent visualization when API is down
• Proof timestamps and agent creation dates reflect actual on-chain activity

Resources

• Giza GitHub
• Giza Documentation
• ZKML Explained