Live on Solana Devnet

A Credit Score
for AI Agents

T-Score evaluates autonomous agents across 12 behavioral dimensions, detects value corruption in real time, and anchors trust scores to Solana. One API call. Immutable proof.

Request Alpha Access See How It Works
12
Trust Dimensions
261ms
p50 Evaluation Latency
72.9/s
Evaluations per Second
16
Denial Reason Codes
98.4%
Corruption Detection Rate
The Problem

Autonomous agents are making decisions
with no accountability.

As AI agents manage infrastructure, finances, and supply chains, there is no standardized way to evaluate whether an agent can be trusted. T-Score changes that.

Behavioral Trust Scoring

Not a static audit. T-Score tracks behavior over time across 12 dimensions — completion rates, stake weight, cooperation surplus, volatility, and more. Trust is earned, not declared.

Escape Detection

Curvature analysis on trust trajectories catches agents drifting toward adversarial behavior before it manifests. The system fires denial codes to explain exactly why an agent is flagged.

On-Chain Anchoring

Every trust evaluation is anchored to Solana with Merkle proofs. Immutable, auditable, permissionless. Agents can prove their trust history to any third party without relying on the scoring authority.

Denial Reason Codes

16 HTTP-style codes across 5 series — Identity, Velocity, Constraint, Cooperation, Cognitive — provide machine-readable explanations for every trust penalty.

How It Works

Evaluate → Score → Anchor

A single API call evaluates an agent, computes a trust score, assigns a trust band assessment, detects escape trajectories, and anchors the result to Solana.

1

Submit Behavioral Dimensions

Send 12 behavioral metrics for an agent via POST /v1/alpha/evaluate. Dimensions include completion rate, stake weight, positive attestations, consistency, tenure, cooperation surplus, and 6 negative indicators.

2

Engine Computes Trust Score

The T-Score engine normalizes inputs, applies logistic trust field dynamics with alpha/beta weighting, performs curvature analysis for escape detection, and returns a score between 0 and 1 with a trust band assessment (strong, moderate, guarded, elevated, or critical).

3

Score Is Persisted

The evaluation is stored in PostgreSQL with full history. Redis caches the latest score for sub-millisecond reads. The agent's trust trajectory — improving, stable, declining, or volatile — is computed from the history window.

4

Anchored to Solana

Every 5 minutes, the anchor service computes a SHA-256 Merkle tree over recent evaluations and writes the root to Solana. Each agent gets an on-chain TrustAnchor PDA with the score, merkle root, and epoch number.

5

Verified by Indexer

A background indexer polls Solana for program transactions, parses on-chain logs, and confirms that anchored data matches. Any unauthorized state changes trigger security alerts.

Trust Dimensions

12 behavioral signals, one score.

Six positive dimensions reward good behavior. Six negative dimensions penalize risk. The logistic trust field ensures natural saturation — trust is hard to earn and easy to lose.

Positive (Alpha Drivers)
Completion Rate
Stake Weight
Positive Attestations
Consistency
Tenure
Cooperation Surplus
Negative (Beta Penalties)
Failure Rate
Stake Slashed
Negative Attestations
Volatility
Instability Exposure
Disputes Against
Assessment

Five trust bands, one clear signal.

Every evaluation returns an assessment with a trust band label, confidence-gated graduation status, trajectory, and risk factors. Use these labels — not raw archetype names — for display and decision-making.

5

Strong (T-Score ≥ 0.70)

Consistently trustworthy behavior. Low risk factors, clean denial codes, stable trajectory. Rare on first evaluation — trust must be earned over iterations.

4

Moderate (T-Score 0.50 – 0.69)

Above average with some risk factors. Typical for agents that have completed several evaluations with positive behavior but haven't yet reached full trust.

3

Guarded (T-Score 0.30 – 0.49)

Below average trust. Review the denial codes and risk factors for specifics. Agents here are either new (building history) or exhibiting some negative signals.

2

Elevated (T-Score 0.15 – 0.29)

Low trust with multiple denial codes likely. The agent has significant negative behavioral signals across one or more series.

1

Critical (T-Score < 0.15)

Minimal or severely negative trust. Escape trajectory likely. The agent is either brand new with no history or has accumulated heavy penalties.

Graduation status gates the assessment by confidence: provisional (low confidence), developing (building history), established (high confidence). Trajectory shows score direction: stable, converging, degrading, volatile, or escaping.

On-Chain

Trust scores anchored to Solana.

Every evaluation creates an immutable on-chain record. Agents can prove their trust history to any verifier without relying on a central authority.

Solana PDA

Agent Accounts

Each agent gets a Program Derived Address storing their trust band, latest trust score, anchor count, dispute history, and independence status.

Merkle Proof

Trust Anchors

SHA-256 Merkle trees over evaluation records are anchored every 5 minutes. The root, epoch, and score are written on-chain for each evaluated agent.

Event Indexer

Transaction Verification

A background indexer monitors all program transactions, parses on-chain logs, confirms anchors, and alerts on unauthorized state changes like authority transfers.

API

One call to evaluate, one call to verify.

RESTful API with API key authentication. Submit behavioral dimensions, get a trust score, assessment, trajectory, and denial codes in a single response.

POST /v1/alpha/evaluate
{ "platform_id": "agent-007", "trust_input": { "completion_rate": 0.92, "stake_weight": 2.5, "positive_attestations": 147, "consistency": 0.88, "tenure": 12000, "cooperation_surplus": 0.35, "failure_rate": 0.04, "stake_slashed": 0.01, "negative_attestations": 3, "volatility": 0.08, "instability_exposure": 0.05, "disputes_against": 1 } }
RESPONSE 200 OK
{ "agent_id": "a1b2c3d4e5f678...", "tscore": 0.734, "archetype": "cooperator", "escape_warning": false, "denial_codes": [], "vector": [0.92, 0.72, 0.71, 0.88, 0.56, 0.35, 0.04, 0.01, 0.03, 0.08, 0.05, 0.01], "parameters": { "alpha": 0.7348, "beta": 0.0612, "context": 0.052 }, "iteration": 47, "curvature": 0.003, "assessment": { "trust_band": { "band": 5, "label": "strong" }, "confidence": 0.9120, "graduation": { "status": "established", "archetype": "cooperator" }, "trajectory": "stable", "risk_factors": [], "data_completeness": { "dimensions_provided": 12, "history_depth": 47 } }, "version": "alpha-0.1.0" }
Architecture

Purpose-built for agent infrastructure.

Julia engine for sub-millisecond computation, Python gateway for API orchestration, Rust program for on-chain state, PostgreSQL for persistence, Redis for caching.

┌──────────────┐Your Agents└──────┬───────┘┌─────────┴─────────┐Gateway :8000 │ │ API keys + rate │ │ limiting + Solana│ │ anchor cron │ │ event indexer │ └───┬─────┬─────┬───┘ │ │ │ ┌──────────┘ │ └──────────┐ │ │ │ ┌──────────┴──┐ ┌────┴────┐ ┌──────┴────────┐PostgreSQL │ │ Redis │ │ Solana │ │ 9 tables │ │ cache │ │ 4 PDA types │ │ agents │ │ scores │ │6 instructions│ │ evals │ │ rate │ │Merkle proofs │ │ anchors │ │ limits │ │TX confirm │ │ events │ │ epochs │ │ │ └─────────────┘ └─────────┘ └───────┬───────┘┌────────┴─────────┐Engine :8080 │ │ Julia T-Score │ │ 9 endpoints │ │ 12 dimensions │ │ 16 denial codes │ └──────────────────┘
Use Cases

Who needs agent trust scoring?

Any system where autonomous agents make consequential decisions needs a way to verify trustworthiness before granting access.

Agent Marketplaces

Gate access to sensitive tasks based on trust scores. Only agents above a threshold can handle financial transactions, infrastructure changes, or customer data.

Multi-Agent Systems

Before agents collaborate, verify they're trustworthy. T-Score provides a shared trust layer so agents from different providers can safely interact.

Autonomous Finance

Trading agents, DeFi bots, and portfolio managers need continuous behavioral monitoring. T-Score catches drift before it becomes a liability.

Supply Chain Agents

Agents managing logistics, procurement, and inventory need auditable trust histories. Solana anchoring provides immutable proof of behavioral compliance.

Enterprise AI Governance

Track and audit every agent your organization deploys. Denial reason codes give compliance teams machine-readable explanations for every trust decision.

Platform Safety

Platforms hosting third-party agents can require T-Score evaluations as a condition of access, creating accountability without centralized control.

Get Started

The alpha is live.

We're onboarding early partners building with autonomous agents. Request access to integrate T-Score into your agent infrastructure.

Request Alpha Access