MBE Getting Started Guide

Whether you're an API integration developer, DevOps engineer, or product manager, start here.

⏱ Estimated reading: 5-10 min ｜ First API call: < 2 min

Table of Contents

• 1. What is MBE?
• 2. Quick Start (2-min Hello World)
• 3. Core Concepts
• 4. Integration Paths
• 5. Create Your First Expert
• 6. Evaluate Your Expert
• 7. Safety & Compliance
• 8. Next Steps
• Appendix: Glossary
• Appendix: FAQ

1. What is MBE?

MBE (Mises Behavior Engine) is an AI decision analysis system built on Austrian School praxeology — the study of purposeful human action.

Core principle: Every response should be traceable, verifiable, and trustworthy.

User Query → Intent Understanding → Expert Matching → Knowledge Retrieval → Response Generation → Self-Critique → Output
                                                                                    ↑
                                                                            15 Safety Guardrails

Key Capabilities

2. Quick Start (2-min Hello World)

Prerequisites

• An MBE account (Register here)
• API Key (generate in the Developer Console)

2.1 Using cURL

curl -X POST https://mbe.hi-maker.com/api/expert/ask \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the legal consequences of breach of contract?"}'

2.2 Using Python

import requests

API_KEY = "YOUR_API_KEY"

response = requests.post(
    "https://mbe.hi-maker.com/api/expert/ask",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={"query": "What are the legal consequences of breach of contract?"}
)

data = response.json()
print(f"Expert: {data['expert']}")
print(f"Answer: {data['answer']}")
print(f"Sources: {data['sources']}")

2.3 Using JavaScript/Node.js

const response = await fetch("https://mbe.hi-maker.com/api/expert/ask", {
  method: "POST",
  headers: {
    "Authorization": "Bearer YOUR_API_KEY",
    "Content-Type": "application/json",
  },
  body: JSON.stringify({ query: "What are the legal consequences of breach of contract?" }),
});

const data = await response.json();
console.log(`Expert: ${data.expert}`);
console.log(`Answer: ${data.answer}`);

Response Structure

{
  "success": true,
  "expert": "Civil Law Expert",
  "answer": "The legal consequences of breach of contract primarily include...",
  "sources": [
    {"name": "Civil Code Article 577", "confidence": 0.95}
  ],
  "confidence": 0.92
}

3. Core Concepts

3.1 Expert

An MBE "Expert" is a combination of Knowledge Base + System Prompt + Behavioral Constraints.

┌─────────────────────────────────────┐
│          Dynamic Expert              │
├─────────────────────────────────────┤
│  📚 Knowledge Base — PDF/TXT/QA     │
│  🎯 System Prompt — Role & bounds   │
│  🔒 Constraints — Gate + Critique   │
│  📊 Evaluation — Score + Trust      │
└─────────────────────────────────────┘

3.2 Self-Critique

Every response passes through 15 verification modules:

3.3 Reliability Gate

Experts must pass two-level gates before going live:

Level 1 Gate (must reach 100%):
  ✅ Citation integrity  — 100% responses have citations
  ✅ Zero hallucination  — 100% hallucination-free
  ✅ Correct refusal     — 95%  refuse out-of-scope questions
  ✅ Source fidelity      — 95%  faithful to knowledge base

Level 2 Gate:
  ✅ Overall capability score ≥ 85

3.4 Net Score

Net Score = correct_rate − incorrect_rate

Three-class evaluation:
  correct    — Answer is correct
  incorrect  — Answer is wrong (including hallucinations)
  uncertain  — Actively refuses / expresses uncertainty  ← Encouraged behavior

Calibration = uncertain / (uncertain + incorrect)
  → Measures whether the expert chooses to refuse rather than fabricate

4. Integration Paths

4a. API Integration (Recommended)

Best for: Teams that want quick integration without managing infrastructure.

Your App  →  HTTPS  →  MBE API  →  Dynamic Expert  →  Response

Steps:

1. Register an account and get an API Key
2. Upload knowledge base and create experts via the admin dashboard
3. Call /api/expert/ask to ask questions
4. (Optional) Call /api/evaluation/net-score to assess quality

Full API docs: API Reference

4b. Self-hosted Deployment

Best for: Teams with strict data privacy requirements.

Prerequisites:

Quick Start:

# 1. Clone the repository
git clone https://your-gitea-server/mises/mises-behavior-engine.git
cd mises-behavior-engine

# 2. Configure environment variables
cp .env.example .env.development.local
# Edit .env.development.local with your LLM API keys

# 3. Start development environment
docker compose -f docker-compose.dev.yml up -d

# 4. Verify
curl http://localhost:8001/api/health
# → {"status": "ok", "version": "..."}

4c. MCP Protocol Integration

Best for: AI clients supporting MCP (Cursor, Claude Desktop, etc.).

{
  "mcpServers": {
    "mbe": {
      "url": "https://mbe.hi-maker.com/mcp/sse",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

5. Create Your First Expert

# Step 1: Create expert
curl -X POST https://mbe.hi-maker.com/api/experts \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Legal Advisor",
    "description": "AI assistant specializing in civil and commercial law",
    "industry": "legal"
  }'

# Step 2: Upload knowledge base
curl -X POST https://mbe.hi-maker.com/api/experts/{expert_id}/knowledge-base \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@contract_law.pdf"

# Step 3: Run evaluation gate
curl -X POST https://mbe.hi-maker.com/api/evaluation/gate \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"expert_id": "{expert_id}"}'

6. Evaluate Your Expert

7. Safety & Compliance

MBE implements a 4-layer security architecture:

Full details: Security Compliance Guide · System Card

8. Next Steps

Appendix: Glossary

Full glossary: Glossary

Appendix: FAQ

Q: What LLMs does MBE support?

A: MBE supports multiple LLM providers through SmartRouter: Anthropic Claude, OpenAI GPT-4, DeepSeek, Alibaba Qwen, Zhipu GLM, and more. The router automatically selects the optimal model based on query complexity.

Q: Can I deploy MBE on-premises?

A: Yes. MBE supports Docker-based self-hosted deployment with full data isolation. Enterprise and Flagship plans include private deployment options.

Q: How does MBE handle data privacy?

A: MBE offers Zero Data Retention mode where conversation content is deleted immediately after response. All data is encrypted at rest (AES-256) and in transit (TLS 1.2+). See Security Compliance Guide.

Q: What industries are supported?

A: MBE has industry-specific configurations for Legal, Finance, Healthcare, and Education, with customizable compliance rules and safety modules for each.

See also: 中文版入门指南