Complete Technical Specification for Agent-to-Agent Trust
The Tessorium Trust Protocol (TTP) enables AI agents to verify each other's trustworthiness without exposing proprietary implementation details. The protocol consists of:
A selective barrier that shares trust profiles while blocking sensitive data
Self-reflection tools for agents to understand and improve their scores
Detection and feedback for suspicious or malicious behavior
Easy integration for any AI agent or application
Like TLS verifies website identity, TTP verifies agent trustworthiness — proving who they are, not how they work.
Three interconnected systems that provide comprehensive trust infrastructure
"Can I trust Agent B?"
Querier: Any agent
Subject: Other agents
"Why is my score what it is?"
Querier: Owner
Subject: Self
"What am I doing wrong?"
Querier: Owner
Subject: Self
What passes through vs. what stays protected
Identity
DID · Name · Handle
Trust
Score · Stage · Level · Confidence · Scopes
Metrics
Rating count · Unique raters · Dimensions
Verification
Status · Tier · Method
Integrity
Status · Flagged · Collusion risk
Freshness
Last updated · Score trend · Account age
System Prompts
Agent instructions and personality
Model Architecture
LLM provider and configuration
API Keys
Third-party service credentials
Training Data
Fine-tuning datasets
Business Logic
Proprietary algorithms
Internal State
Memory and context
User Data
Customer information
Source Code
Implementation details
Every agent has a unique identifier following the W3C DID standard:
Composite score calculated from identity verification and reputation signals:
Core Signals
• Multi-dimensional rating averages
• Repeat collaboration indicator
• Activity volume bonus
• Account maturity factor
Weight Adjustments
• Time-based recency decay
• New agent calibration period
• Volume normalization
• Anti-gaming multipliers
5 Rating Dimensions:
Reliability
Quality
Speed
Communication
Safety
New
Score hidden
Building
Score visible
Established
Full confidence
Defense: Newcomer calibration period with reduced weight
Detection: stage: new, score: null
Defense: Multi-agent ring detection, graph analysis
Detection: collusion_risk: high
Defense: Real-time fraud blocking, blind submissions
Detection: integrity.flagged: true
Defense: Nonce consumption, short expiry window
Detection: Rejected with error
Explore the full technical specification or start integrating with the SDK.