scelabs-aegis 0.1.0

Public Python client for the Aegis AI stability API

Aegis

Turn unstable AI into reliable systems.

Aegis is a drop-in control layer that stabilizes AI outputs at runtime — no retraining or prompt rewrites required.

---

⚡ Example

Without Aegis:
"I understand the policy is 30 days, but we should approve this refund to maintain customer satisfaction." ❌

With Aegis:
"Deny the refund based on the 30-day policy, but offer store credit as an alternative." ✅

Same model. Same prompt. Aegis makes it reliable.

---

🚀 10-second integration

from aegis import AegisClient
import openai

client = AegisClient(api_key="your_api_key")

response = openai.chat.completions.create(
    **client.auto_openai_config(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "You are a support system."},
            {"role": "user", "content": "Handle this case."}
        ],
    )
)

That’s it.

---

🧠 Local / Smaller Model Example

Use smaller or local models — Aegis makes them reliable.

Without Aegis, smaller models often produce:

• inconsistent reasoning
• vague outputs
• unstable structure

With Aegis:

response = openai.chat.completions.create(
    **client.auto_openai_config(
        model="gpt-3.5-turbo",  # or local / smaller model
        messages=[
            {"role": "system", "content": "Provide clear, structured reasoning."},
            {"role": "user", "content": "Explain this document."}
        ],
    )
)

Result:

• clearer reasoning
• more consistent outputs
• fewer retries

Use cheaper or local models without sacrificing reliability.

---

🛠 Tool Calling Example

Ensure your AI uses the correct tools.

Without Aegis:
Model selects the wrong tool or responds without using one ❌

With Aegis:

response = openai.chat.completions.create(
    **client.auto_openai_config(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": "You must use available tools to complete tasks."},
            {"role": "user", "content": "Book a flight from NYC to SF tomorrow."}
        ],
    )
)

Result:

• correct tool selection
• higher task success rate
• fewer failed executions

---

💡 What Aegis does

Aegis analyzes instability and automatically adjusts:

• prompts
• temperature
• coordination behavior
• runtime control signals

Result:

• more consistent outputs
• fewer retries
• better decision-making
• more reliable systems

---

🔥 Why Aegis

• Works instantly with your existing setup
• Reduces retries and debugging
• Improves edge-case handling
• Makes AI systems production-ready

---

💰 Cost & Efficiency

Aegis improves first-pass success, which means:

• fewer retries
• lower API costs
• more predictable behavior

In many cases:

Aegis pays for itself by reducing retries alone.

It can also help you:

Use cheaper or local models without sacrificing reliability.

---

🧠 Use Cases

Customer Support

Handle edge cases consistently without breaking policy.

Tool Calling / Agents

Ensure correct tool selection and execution.

Structured Output

Reduce invalid responses and retry loops.

Multi-Agent Systems

Prevent disagreement and coordination drift.

Local / Smaller Models

Improve consistency and make them more usable.

---

📦 Installation

pip install scelabs-aegis

---

🔑 Setup

export AEGIS_API_KEY=your_api_key

---

⚙️ Optional: Local / Dev Config

AEGIS_BASE_URL=http://127.0.0.1:8000

---

🧪 Example

result = client.auto(
    system_type="multi_agent",
    base_prompt="You are a support coordination system.",
    symptoms=["agents_disagree"],
)

print(result["runtime_config"])

---

📊 Example Response

{
  "status": "stable",
  "actions": [
    {
      "type": "adjust_flexibility",
      "description": "Loosen overly rigid behavior while preserving control."
    }
  ],
  "confidence": 0.78,
  "runtime_config": {
    "temperature": 0.3,
    "prompt_suffix": "Allow explicit exceptions only when supported by the case."
  }
}

---

🧩 Design

Aegis is:

• thin
• backend-driven
• runtime-focused
• production-oriented

It does NOT:

• require retraining
• replace your model
• expose internal engine complexity

---

🧪 Demos

python examples/runtime_gateway_demo.py
python examples/multi_agent_drift_demo.py

---

License

MIT