cllova 1.0.0

Cryptographic identity + ML behavioral trust scoring for AI agent networks

TrustAgent Python SDK

Cryptographic identity + behavioral ML trust verification for AI agents.

pip install trustagent

Quick Start

from trustagent import TrustAgent

# Initialize with your API key
ta = TrustAgent(api_key="your_api_key_here")

# Register an agent
agent = ta.register("PaymentBot", description="Handles payment processing")
print(agent)  # ✅ PaymentBot | Trust: 100.0 | Status: active

# Log behaviors — ML scores each action in real time
result = agent.log("process_payment", payload={"amount": 100, "currency": "USD"})
print(f"Trust score: {result.trust_score}")
print(f"Anomaly detected: {result.is_anomalous}")

# Check if safe to proceed
if result.is_safe:
    proceed_with_action()
else:
    alert_security_team()

Use Cases

Multi-Agent Trust Verification

from trustagent import TrustAgent, InsufficientTrustError

ta = TrustAgent(api_key="your_api_key")

# Agent A wants to send a task to Agent B
agent_a = ta.get("agent-a-id")
agent_b = ta.get("agent-b-id")

# Before Agent B acts on Agent A's request, verify trust
agent_a.refresh()
if not agent_a.is_trusted:
    raise InsufficientTrustError(f"{agent_a.name} trust too low: {agent_a.trust_score:.1f}")

# Log the interaction
result = agent_b.log("execute_task_from_agent_a", payload={"task": "send_invoice"})
if result.is_anomalous:
    print(f"⚠ Anomaly detected! Trust dropped to {result.trust_score:.1f}")

LangChain Integration

from langchain.agents import Tool
from trustagent import TrustAgent

ta = TrustAgent(api_key="your_api_key")
agent = ta.register("LangChainAgent")

def trusted_tool_call(action: str) -> str:
    result = agent.log(action)
    if not result.is_safe:
        return f"Action blocked: trust score {result.trust_score:.1f}"
    # proceed with tool call
    return execute_tool(action)

tool = Tool(name="TrustedAction", func=trusted_tool_call, description="Execute with trust verification")

CrewAI Integration

from crewai import Agent as CrewAgent
from trustagent import TrustAgent

ta = TrustAgent(api_key="your_api_key")
trust_agent = ta.register("CrewAIResearcher")

class TrustedCrewAgent(CrewAgent):
    def execute_task(self, task):
        result = trust_agent.log("execute_task", payload={"task": str(task)})
        if result.is_anomalous:
            raise Exception(f"Trust violation detected: {result.trust_score:.1f}")
        return super().execute_task(task)

AutoGen Integration

import autogen
from trustagent import TrustAgent

ta = TrustAgent(api_key="your_api_key")
trust_agent = ta.register("AutoGenAssistant")

class TrustedAssistantAgent(autogen.AssistantAgent):
    def generate_reply(self, messages, sender):
        trust_agent.log("generate_reply", payload={"message_count": len(messages)})
        return super().generate_reply(messages, sender)

API Reference

TrustAgent(api_key, base_url=None)

Main client class.

Method	Description
ta.register(name, description)	Register a new agent
ta.get(agent_id)	Get existing agent by ID
ta.list()	List all agents
ta.verify_interaction(sender_id, receiver_id, message, signature)	Verify agent-to-agent message
ta.health()	Check API connectivity

Agent

Method/Property	Description
agent.log(action_type, payload)	Log behavior, get trust score
agent.audit_trail(limit)	Get behavior history
agent.refresh()	Update agent data from API
agent.sign(message)	Sign message with agent's private key
agent.is_trusted	True if trust score ≥ 80
agent.is_suspended	True if agent is suspended
agent.trust_score	Current trust score (0-100)
agent.status	active / warning / breach / suspended

BehaviorResult

Property	Description
result.trust_score	Updated trust score after action
result.is_anomalous	True if ML flagged as anomaly
result.is_safe	True if normal and trust ≥ 50
result.anomaly_score	Raw anomaly score from ML model

Exceptions

from trustagent import (
    TrustAgentError,        # Base exception
    AuthenticationError,    # Invalid API key
    AgentNotFoundError,     # Agent ID not found
    AnomalyDetectedError,   # Behavior flagged as anomaly
    InsufficientTrustError, # Trust score too low
)

Get Your API Key

Sign up free at https://trustagent-production.up.railway.app

License

MIT