anthropic-omega 0.1.0

Anthropic Claude SDK with OmegaEngine governance - AI safety and compliance

anthropic-omega

Anthropic Claude Integration with OmegaEngine Governance

Documentation • Playground • Issues

---

✨ Features

• 🤖 Drop-in Replacement — Same API as official Anthropic SDK
• 🛡️ Pre-flight Checks — Validate prompts before sending to Claude
• 📊 Response Governance — Filter outputs against policies
• 📈 Full Audit Trail — Log every Claude interaction
• ⚡ Streaming Support — Governed streaming responses
• 🔄 Constitutional AI — Add policy constraints to any Claude call

---

📦 Installation

Python

pip install anthropic-omega

---

🚀 Quick Start (Python)

from anthropic_omega import Anthropic

# Drop-in replacement for Anthropic client
client = Anthropic(
    anthropic_api_key="your-anthropic-key",
    omega_api_key="your-omega-key",
    policy_id="claude_policy",
)

# Same API, but governed
message = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Help me write a marketing email"}
    ]
)

print(message.content[0].text)

# Access governance metadata
print(message.omega_decision)      # "APPROVE"
print(message.omega_audit_id)      # "aud_abc123"

---

---

🛡️ Pre-flight Validation

Block unsafe prompts before they reach Claude:

client = Anthropic(
    anthropic_api_key="...",
    omega_api_key="...",
    preflight_check=True,  # Validate prompts first
    block_on_deny=True,    # Don't send if denied
)

try:
    # This will be blocked if policy denies
    message = client.messages.create(
        model="claude-3-5-sonnet-20241022",
        messages=[{"role": "user", "content": "...potentially unsafe..."}]
    )
except OmegaBlockedError as e:
    print(f"Blocked: {e.reason}")
    print(f"Policy tags: {e.policy_tags}")

---

📡 Governed Streaming

with client.messages.stream(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Write a story"}]
) as stream:
    for text in stream.text_stream:
        print(text, end="", flush=True)

# Final response is still validated
print(f"\nDecision: {stream.omega_decision}")

---

🎯 Use Cases

Use Case	How It Helps
Customer Support Bots	Ensure Claude doesn't share internal policies
Content Generation	Block harmful or off-brand content
Code Assistants	Prevent generation of malicious code
Healthcare Apps	Enforce HIPAA compliance
Finance Apps	Ensure FINRA compliance

---

📊 What Gets Logged

Every Claude call logs:

• ✅ Input messages
• ✅ Output response
• ✅ Token usage
• ✅ Model used
• ✅ Latency
• ✅ Policy evaluation
• ✅ Risk assessment
• ✅ Audit trail ID

---

📄 License

Licensed under the Apache License 2.0.

---

Built with ❤️ by the OmegaEngine team