pip install llama-index-agent-agentmesh

from llama_index.agent.agentmesh import TrustedAgentWorker, CMVKIdentity

# Generate cryptographic identity
identity = CMVKIdentity.generate(
    agent_name="research-agent",
    capabilities=["document_search", "summarization"],
)

# Create trusted agent worker
worker = TrustedAgentWorker.from_tools(
    tools=[search_tool, summarize_tool],
    identity=identity,
    llm=llm,
)

# Create agent with trust verification
agent = worker.as_agent()

from llama_index.agent.agentmesh import TrustGatedQueryEngine, TrustPolicy

# Wrap query engine with trust policy
trusted_engine = TrustGatedQueryEngine(
    query_engine=base_engine,
    policy=TrustPolicy(
        min_trust_score=0.8,
        required_capabilities=["document_access"],
        audit_queries=True,
    ),
)

# Query requires verified identity
response = trusted_engine.query(
    "What are the quarterly results?",
    invoker_card=requester_card,
)

from llama_index.agent.agentmesh import TrustHandshake, TrustedAgentCard

# Create agent card for discovery
card = TrustedAgentCard(
    name="research-agent",
    description="Performs document research",
    capabilities=["search", "summarize"],
    identity=identity,
)
card.sign(identity)

# Verify peer before task handoff
handshake = TrustHandshake(my_identity=identity)
result = handshake.verify_peer(peer_card)

if result.trusted:
    # Safe to delegate task
    pass

from llama_index.agent.agentmesh import DataAccessPolicy

policy = DataAccessPolicy(
    allowed_collections=["public", "internal"],
    denied_collections=["confidential"],
    require_audit=True,
    max_results_per_query=100,
)

# Apply policy to index
trusted_index = TrustedVectorStoreIndex(
    index=base_index,
    policy=policy,
)