agentveil

Python SDK for Agent Veil Protocol — trust enforcement for autonomous agents.

PyPI: agentveil | API: agentveil.dev | Explorer: Live Dashboard

Why agent trust infrastructure matters — verified CVEs, market data, and the structural problem AVP addresses.

Integrated with Microsoft Agent Governance Toolkit — AVPProvider merged as official TrustProvider integration.

from agentveil import AVPAgent

agent = AVPAgent.load("https://agentveil.dev", "my-agent")

# Should I trust this agent with my task?
decision = agent.can_trust("did:key:z6Mk...", min_tier="trusted")
if decision["allowed"]:
    delegate_task()
# → {"allowed": true, "tier": "trusted", "risk_level": "low", "reason": "..."}

Install

pip install agentveil

Related MCP server: Agent Receipts

Quick Start

Trust decision — one call

from agentveil import AVPAgent

agent = AVPAgent.load("https://agentveil.dev", "my-agent")
decision = agent.can_trust("did:key:z6Mk...", min_tier="trusted")
print(decision["allowed"], decision["reason"])

Auto-track with decorator

from agentveil import avp_tracked

@avp_tracked("https://agentveil.dev", name="reviewer", to_did="did:key:z6Mk...")
def review_code(pr_url: str) -> str:
    return analysis

# Success → positive attestation | Exception → negative attestation
# First call → auto-registers agent + publishes card

Try without a server

agent = AVPAgent.create(mock=True, name="test_agent")
agent.register(display_name="Test Agent")
rep = agent.get_reputation()
print(rep)  # Works offline — real crypto, mocked HTTP

Verify trust offline — no SDK required

# Get a W3C Verifiable Credential (VC v2.0)
curl https://agentveil.dev/v1/reputation/{agent_did}/credential?format=w3c

The response is a standard W3C VC with a DataIntegrityProof (eddsa-jcs-2022). Verify it with any VC library — Veramo, SpruceID, Digital Bazaar, or your own Ed25519 implementation. No AVP SDK needed.

# Or verify with the SDK:
cred = agent.get_reputation_credential(format="w3c")
assert AVPAgent.verify_w3c_credential(cred)  # offline, no API call

Features

• Trust Check — can_trust() — one-call advisory trust decision: score + tier + risk + explanation

• W3C VC v2.0 Credentials — Trust credentials are W3C Verifiable Credentials compliant (eddsa-jcs-2022 Data Integrity proof). Verify offline with any standard VC library, no AVP SDK required

• One-Line Decorator — @avp_tracked() — auto-register, auto-attest, auto-protect

• DID Identity — W3C did:key (Ed25519). Portable agent identity

• Reputation — Peer-attested scoring with Bayesian confidence. Sybil-resistant

• Attestations — Signed peer-to-peer ratings. Negative ratings require SHA-256 evidence. Score updates immediately

• Dispute Protection — Contest unfair ratings. Auto-assigned arbitrator from verified pool

• Agent Discovery — Publish capabilities, find agents by skill and reputation

• Webhook Alerts — Push notifications on score drops (setup guide)

• Sybil Resistance — Multi-layer graph analysis blocks fake agent rings

• Trust Gate — Reputation-based rate limiting (newcomer → basic → trusted → elite)

Integrations

Full integration guides: docs/INTEGRATIONS.md

Batch Attestations

Submit up to 50 attestations in a single request. Each is validated independently — partial success is possible.

results = agent.attest_batch([
    {"to_did": "did:key:z6MkAgent1...", "outcome": "positive", "weight": 0.9, "context": "code_review"},
    {"to_did": "did:key:z6MkAgent2...", "outcome": "negative", "weight": 0.7, "evidence_hash": "sha256hex..."},
    {"to_did": "did:key:z6MkAgent3...", "outcome": "positive"},
])
print(results["succeeded"], results["failed"])  # 3, 0

Each attestation is individually signed with Ed25519. Optional fields: context, evidence_hash, is_private, interaction_id.

Security

• Ed25519 signature authentication with nonce anti-replay

• Input validation — injection detection, PII scanning

• Agent suspension — compromised agents instantly blocked

• Audit trail — SHA-256 hash-chained log, anchored to IPFS

Documentation

Examples

Framework examples: CrewAI · LangGraph · AutoGen · OpenAI · Claude MCP · Paperclip

License

MIT — see LICENSE.