AI Integration (ASI)
5. AI Integration (ASI)
5.1 Multi-Agent Architecture
Creo uses a distributed agent marketplace for intelligent code analysis:
Repository Analysis Flow
User submits repository URL to main agent
Main agent discovers available uAgents specializing in code review
Custom Langchain Code Analyzer identified alongside other agents
Main agent selects 3 optimal uAgents for analysis
Each uAgent analyzes the repository using their methodology
Main agent synthesizes all responses into final recommendation
Pull Request Review Flow
Specialized PR Review Agent handles pull request analysis
Uses MeTTa for enhanced reasoning capabilities
Integrates with distributed agent discovery system
Provides detailed code review insights and suggestions
5.2 Verifiable AI: Two Approaches
Current: Optimistic Verification (Fetch.ai)
AI outputs assumed valid by default
Agent identities and reputation scores stored on-chain
Community can challenge suspicious results
Challenges reduce agent reputation (similar to optimistic rollup slashing)
Malicious agents lose selection priority over time
Limitations:
Requires trusting that challengers will notice malicious outputs
Verification is social, not cryptographic
Challenge period introduces delays
Future: Cryptographic Verification (0G Compute + TEE)
All AI computations run inside Trusted Execution Environments
TEE generates cryptographic attestation proving:
Exact code executed
Exact inputs provided
Exact outputs produced
No tampering occurred
Anyone can verify proof on-chain instantly
No trust assumptions, no social coordination required
Comparison:
5.3 MeTTa Integration
MeTTa enables advanced symbolic reasoning for code analysis:
Organizes knowledge as logical graphs
Supports pattern matching and symbolic reasoning
Allows agents to query structured repository information
Enables efficient reasoning on code review tasks
Combines structured knowledge with LLM capabilities
5.4 AI Credits System
How It Works:
Organizations add AI credits (ETH) to their account
Each AI computation costs 0.0000001 ETH
Credits transferred to AI agent address
Auto-deducted when AI creates issues or grades submissions
Use Cases:
Automated issue generation from repository analysis
PR grading and confidence scoring
Feature suggestion based on market analysis
Multi-agent code review
Last updated