sqube-agent-guard

Sqube Agent Control — open-source execution authorization for AI agents.

Evaluate deterministic policy (ALLOW / BLOCK / REQUIRE_APPROVAL), optional human approval, and a tamper-evident execution ledger. The SDK is application-level; bypassing the guard is out of scope for v1.

Install

pip install sqube-agent-guard

Optional extras:

pip install "sqube-agent-guard[control-plane]"   # local API + dashboard
pip install "sqube-agent-guard[otel]"            # OpenTelemetry bridge

Requires Python 3.10+.

Quick start

from sqube_agent_guard import Decision, ExecutionGuard

def policy(action, resource, agent_id, **ctx):
    if action == "email.send":
        return Decision.REQUIRE_APPROVAL
    return Decision.ALLOW

guard = ExecutionGuard(policy=policy)

@guard.wrap_action(action="email.send", resource="user@example.com")
def send_email():
    ...

send_email()

Control plane (optional)

sqube-agent-guard serve --data-dir .sqube

Open http://127.0.0.1:8080/setup on first run to create an administrator account.

Documentation

https://sqube-groups.github.io/sqube-agent-control/docs/intro

License

Apache-2.0 — Copyright © 2026 Sqube Groups