z-grc 0.0.26

Governance Engine by zeb labs

Governance, Risk, and Control Engine for LLMs

Built by Zeb Labs

---

Enterprise-grade governance engine for Large Language Model applications. Provides automatic interception, policy enforcement, quota management, and comprehensive observability across multiple LLM providers with zero code changes.

Installation

uv add z-grc

Or with auto-instrumentation:

uv add z-grc[auto-instrument]

Quick Start

import zgrc
import boto3
import json

# Initialize GRC
zgrc.init(api_key="your-zgrc-api-key")

# Use your LLM SDK normally - GRC handles everything
client = boto3.client("bedrock-runtime", region_name="us-east-1")

response = client.invoke_model(
    modelId="us.anthropic.claude-sonnet-4-5-20250929-v1:0",
    body=json.dumps({
        "anthropic_version": "bedrock-2023-05-31",
        "max_tokens": 1024,
        "messages": [{"role": "user", "content": "Hello!"}]
    })
)

# Z-GRC automatically:
# - Validates quota before requests
# - Tracks token usage
# - Enforces policies
# - Sends telemetry (traces, metrics, logs)

Features

Zero-Code Integration

Drop-in solution requiring only zgrc.init(). Works with existing code without modifications.

Auto-Discovery

Automatically detects and intercepts installed LLM SDKs:

• AWS Bedrock (boto3)
• Anthropic (coming soon)
• OpenAI (coming soon)
• Azure OpenAI (coming soon)

Policy Enforcement

Real-time quota validation and cost limit enforcement. Blocks requests when quota is exceeded.

from zgrc.utils import QuotaExceededException

try:
    response = client.invoke_model(...)
except QuotaExceededException as e:
    print(f"Quota exceeded: ${e.used:.4f} used, ${e.remaining:.4f} remaining")

Auto-Instrumentation

Optional automatic instrumentation for HTTP clients, web frameworks, databases, and more:

zgrc.init(
    api_key="your-zgrc-api-key",
    auto_instrument=True,
    app_name="my-app",
    environment="production"
)

Framework Agnostic

Works with vanilla SDKs and popular frameworks:

# PydanticAI
from pydantic_ai import Agent
agent = Agent("bedrock")
result = await agent.run("Your prompt")

# LangChain
from langchain_aws import ChatBedrock
llm = ChatBedrock(model_id="...")
response = llm.invoke("Your prompt")

# Strands Agents
from strands_agents import Agent
agent = Agent(provider="bedrock")
response = agent.execute("Your prompt")

Streaming Support

Fully supports streaming responses with automatic token tracking:

response = client.converse_stream(
    modelId="...",
    messages=[{"role": "user", "content": [{"text": "Tell me a story"}]}]
)

for event in response["stream"]:
    if "contentBlockDelta" in event:
        print(event["contentBlockDelta"]["delta"]["text"], end="")

Configuration

zgrc.init(
    api_key: str,                  # Your Z-GRC API key (required)
    auto_instrument: bool = False, # Enable auto-instrumentation
    app_name: str = None,          # Application name for telemetry
    environment: str = None,       # Environment (dev/staging/prod)
    log_level: int = logging.ERROR # Z-GRC internal log level
)

Proxy Mode (Claude Code CLI)

For environments where code modification isn't possible (like Claude Code CLI), use the standalone proxy:

Quick Start

Background Mode (Recommended):

# Set proxy environment variables automatically
eval $(z-grc-proxy --api-key=your-key -d)

# Now run Claude Code - it will use the proxy
claude

Foreground Mode:

# Run proxy in foreground (shows logs, blocks terminal)
z-grc-proxy --api-key=your-key

# In another terminal, set env vars manually:
export HTTPS_PROXY=http://127.0.0.1:8080
export NODE_EXTRA_CA_CERTS=~/.mitmproxy/mitmproxy-ca-cert.pem

claude

Proxy Commands

# Start in background (auto port detection)
eval $(z-grc-proxy --api-key=your-key -d)

# Start on specific port
eval $(z-grc-proxy --api-key=your-key --port=8085 -d)

# Check active proxy sessions
z-grc-proxy --status

# Kill all proxy servers
z-grc-proxy --kill-all

# Verbose logging
eval $(z-grc-proxy --api-key=your-key -d --verbose)

How It Works

1. Automatic Port Detection: Finds available port (8080-8090)
2. Session Management: Reuses existing proxy for same API key
3. mitmproxy Certificates: Auto-generated in ~/.mitmproxy/ on first run
4. Platform Independent: Works on macOS, Linux, Windows

Building Executables

Build standalone proxy binary with PyInstaller:

# Current platform only
make grpc-proxy-build

Output: dist/z-grc-proxy

Test Binary

# Background mode
eval $(./dist/z-grc-proxy --api-key=your-key -d)

# Foreground mode
./dist/z-grc-proxy --api-key=your-key

Installing Executor

macOS / Linux

curl -fsSL https://raw.githubusercontent.com/zeb-ai/z-grc/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/zeb-ai/z-grc/main/install.ps1 | iex