langchain-agentmesh 0.1.0

AgentMesh trust layer integration for LangChain - cryptographic identity and trust-gated tool execution

LangChain AgentMesh Integration

Cryptographic identity verification and trust-gated tool execution for LangChain agents.

Installation

pip install langchain-agentmesh

Features

• CMVKIdentity: Ed25519-based cryptographic identity for agents
• TrustGatedTool: Wrap any tool with trust requirements
• TrustedToolExecutor: Execute tools with automatic verification
• TrustCallbackHandler: Monitor trust events during chain execution
• TrustHandshake: Verify peer agents before collaboration
• DelegationChain: Hierarchical capability delegation

Quick Start

from langchain_agentmesh import CMVKIdentity, TrustGatedTool, TrustedToolExecutor

# Generate agent identity
identity = CMVKIdentity.generate('research-agent', capabilities=['search', 'summarize'])

# Wrap a tool with trust requirements
gated_tool = TrustGatedTool(
    tool=search_tool,
    required_capabilities=['search'],
    min_trust_score=0.8
)

# Execute with verification
executor = TrustedToolExecutor(identity=identity)
result = executor.invoke(gated_tool, 'query')

Use Cases

Multi-Agent Trust Verification

from langchain_agentmesh import TrustHandshake

# Create handshake for peer verification
handshake = TrustHandshake(my_identity)

# Verify peer before collaboration
result = await handshake.verify_peer(peer_identity)
if result.trusted:
    # Safe to delegate task
    response = await peer_agent.invoke(task)

Trust-Gated Tool Execution

from langchain_agentmesh import TrustGatedTool

# Sensitive tool requiring high trust
code_execution_tool = TrustGatedTool(
    tool=python_repl,
    required_capabilities=['code:execute'],
    min_trust_score=0.9,
    audit_logging=True
)

# Only trusted agents can use this tool
result = executor.invoke(code_execution_tool, code)

Callback Integration

from langchain_agentmesh import TrustCallbackHandler

# Monitor trust events
callback = TrustCallbackHandler(
    on_verification=lambda r: print(f"Verified: {r.peer_did}"),
    on_violation=lambda v: alert(f"Violation: {v}")
)

agent = create_agent(callbacks=[callback])

Configuration

Parameter	Default	Description
min_trust_score	0.5	Minimum trust score required
required_capabilities	[]	Required capability list
audit_logging	False	Enable audit trail
cache_ttl	900	Verification cache TTL (seconds)

Related

• AgentMesh - Core trust mesh platform
• Agent-OS - Governance kernel

License

MIT