viatorisai 1.0.0

Python SDK for the Agent Trust Protocol — verifiable identity, cryptographic receipts, and reputation scoring for AI agents

Viatoris Python SDK

Trust infrastructure for AI agents — cryptographic identity, signed receipts, and audit trails.

Built on W3C DIDs and Ed25519 signatures. Works with any agent framework: LangChain, CrewAI, AutoGen, or your own.

Install

pip install viatorisai

Requires Python 3.10+. Only dependency is PyNaCl for Ed25519 signing.

Quick Start

Get an API key from dashboard.viatoris.ai, then:

from agent_trust import AgentTrust

client = AgentTrust(org_api_key="at_live_...")

# Create an agent — generates an Ed25519 key pair locally
passport = client.create_passport(
    agent_id="scout-1",
    controller_did="did:web:example.com",
    metadata={
        "name": "Scout",
        "description": "Research agent",
        "version": "1.0.0",
        "agentType": "autonomous",
        "capabilities": ["search"],
    },
)
# IMPORTANT: save passport.private_key_jwk securely — only returned once.

# Sign and submit a receipt
signer = client.create_signer(passport)
signer.record_action(
    action="web-search",
    capability="research",
    inputs={"query": "AI safety papers"},
    outputs={"results": 12},
    result="success",
)

Decorator pattern

Wrap any function with @client.track to automatically emit a signed receipt for every call. Inputs and outputs are captured from the function signature, and submission is fire-and-forget so it never adds latency.

@client.track(passport, capability="research")
def search(query: str):
    return {"results": do_search(query)}

# Each call to search(...) now produces a signed receipt in the background.
search("AI safety papers")

If the wrapped function raises, the receipt is recorded with result="failure" and the exception is re-raised.

Fire-and-forget receipts

For inline call sites where you don't want a decorator, use record_action_async to submit on a background thread:

signer.record_action_async(
    action="web-search",
    capability="research",
    inputs={"query": "AI safety papers"},
    outputs={"results": 12},
    result="success",
)
# Returns immediately. Errors are logged, never raised.

What you get

• Cryptographic identity — every agent has a W3C DID and Ed25519 key pair. Private keys never leave your process.
• Signed receipts — every action produces a tamper-proof, timestamped record. Inputs and outputs are SHA-256 hashed before signing, so payload contents stay private.
• Public verification — anyone can verify an agent's identity and receipts via https://api.viatoris.ai/v1/verify/<did>. No API key required.
• Reputation scoring — Wilson score confidence intervals computed from your verified receipt history.
• Compliance profiles — attach HIPAA, SOC2, GDPR, or custom policy declarations to your agents.

Documentation

Full guide at viatoris.ai/quickstart.

Questions? Email [e-mail].

License

MIT