Python SDK

Govern any Python AI agent — one import, one line of config.

The node9 Python package wraps your agent's tool calls with the same interception layer as the CLI proxy — without requiring hook wiring or MCP. Every tool call is sent to the Node9 firewall before execution, giving you audit logs, approval flows, and shield enforcement for any Anthropic or custom Python agent.

Install

pip

pip install node9

Usage

Configure your agent

from node9 import configure

# Call once at agent startup — before any tool calls
configure(
    agent_name="my-agent",      # identifies this agent in audit logs
    policy="require_approval",  # audit | require_approval | block_on_rules
)

# Your existing Anthropic agent code runs unchanged
client = anthropic.Anthropic()
response = client.messages.create(...)

Policy modes

audit

Every tool call is logged. Nothing is blocked. Use this for observability before enforcing rules.

require_approvalRecommended

Dangerous tool calls pause and wait for human approval via the Node9 dashboard or Slack before executing.

block_on_rules

Blocks actions that match active shields and smart rules. Everything else is auto-allowed.

Example — CI Code Review Agent

The governed-agent repo contains a complete reference implementation: a CI code review agent that fixes failing tests, posts a review comment, and runs a security scan — all governed by Node9.

ci-code-review-fixer

# governed-agent/ci-code-review-fixer/agent.py
from node9 import configure

configure(agent_name="ci-code-review", policy="require_approval")

# Agent runs autonomously in CI — Node9 intercepts tool calls
# and routes dangerous ones for human approval before they execute

Requirements

Python 3.9+. Requires a running Node9 daemon (node9 daemon start) or a Node9 API key for cloud mode.

Python SDK · node9 documentation

Python SDK

Install

Usage

Policy modes

Example — CI Code Review Agent