        Home/Developers
      

Governance
in a few lines

Wrap any agent in a single MOSS init call. Every action it takes is checked against policy before it leaves the process and signed with ML-DSA-44 as tamper-evident evidence. Your code runs the way it always did.

Request a demo See integrations

governed_agent.py

from moss_agent_sdk import init

agent = init(
    api_key="moss_live_...",
    agent_subject="moss:agent:support-triage",
    declared_behavior={
        "allowed_actions": ["http_get", "email_send"],
        "allowed_destinations": ["https://api.crm.example"],
    },
)

# governed egress — denied calls never open a socket
resp = agent.get("https://api.crm.example/tickets")

# explicit decision point, signed as evidence
with agent.action("email_send"):
    send_reply(resp)

agent.kill("done")  # flush + final signed record + exit

QUICKSTART

How integration works

{{ s.num }}

TWO WAYS TO INTEGRATE

Sign an output, or govern the whole agent

Signing SDK

moss-sdk

Sign and verify any agent output. Verification runs offline; public keys are fetched once and cached locally.

from moss import sign, verify

env = sign(output, agent_id="order-bot").envelope
verify(env).valid   # True — offline

Agent Runtime SDK

moss-agent-sdk

Wrap an entire agent. Policy runs on every egress, backed by a heartbeat lease, an offline revocation cache, and a fail-closed kill switch.

agent = init(agent_subject="moss:agent:bot")

agent.get(url)        # governed egress
agent.kill("done")    # final signed record

INTEGRATIONS

Drops into your agent framework

{{ i.install }}

MOSS also signs agentic commerce. UCP (Google/Shopify) and ACP (OpenAI/Stripe) checkout flows carry provenance for which agent performed each action.

SDKs

Native in your language