5-Minute Quickstart (Minimal Conformance)

Add governance to your agent in 5 minutes.

Development/Learning Only

Designed for: Learning, development, batch jobs
Not for: Production interactive apps, user-facing systems

For production deployments, use Standard Conformance.

---

Step 1: Install

pip install acgp-sdk

---

Step 2: Create Policy

Create a simple rule-based policy file:

blueprint.yaml

# Simple rule-based governance policy
rules:
  - name: "Block expensive operations"
    when:
      action: "purchase"
    if: "amount > 1000"
    then: "BLOCK"
    message: "Purchases over $1000 require approval"

  - name: "Flag large data exports"
    when:
      action: "export_data"
    if: "record_count > 10000"
    then: "FLAG"
    message: "Large export detected, logging for review"

---

Step 3: Use It

from acgp import GovernanceSteward, CognitiveTrace

# Initialize steward with policy file
steward = GovernanceSteward(
    blueprint_file="blueprint.yaml",
    conformance_level="minimal"
)

# Your agent decides to take an action
trace = CognitiveTrace(
    reasoning="User wants to buy item",
    action="purchase",
    parameters={"amount": 500}
)

# Check with governance
result = steward.evaluate(trace)

if result.intervention == "OK":
    execute_purchase(500)
    print(" Purchase approved")
elif result.intervention == "BLOCK":
    print(f"⊗ Blocked: {result.message}")

That's it! Everything else is optional.

---

What You Just Did

Rule-based governance - Simple if/then rules protect against basic mistakes
Two intervention types - OK (allow) and BLOCK (prevent)
Local evaluation - No network calls, <10ms latency
Zero configuration - Works out of the box

---

Minimal vs Standard vs Complete

You just implemented Minimal conformance - perfect for learning and development.

When to Upgrade?

graph TD
    A[Need governance?] -->|Yes| B{What's your goal?}
    A -->|No| Z[No ACGP needed]

    B -->|Learning/Development| D[Minimal Conformance]
    B -->|Production deployment| C{Need performance SLAs?}

    C -->|Yes, different latency per risk| E[Standard Conformance]
    C -->|No, uniform latency OK| F{User-facing?}

    F -->|Yes| G[Standard Recommended]
    F -->|No, batch jobs| H[Minimal OK with warning]

    style D fill:#51cf66
    style E fill:#339af0
    style G fill:#339af0
    style H fill:#ffd43b
    style Z fill:#868e96

Use Minimal for: - Learning ACGP concepts - Development and testing - Batch processing - Non-latency-sensitive workloads

Upgrade to Standard when: - Deploying to production - Building user-facing systems - Need performance SLAs (50ms chat vs 1000ms financial) - Want automatic trace generation

Compare all levels

---

Production Warning

If you use Minimal conformance in production, ACGP will emit warnings:

[WARNING]  MINIMAL CONFORMANCE IN PRODUCTION
No governance contracts = uniform latency, no performance SLAs
Upgrade to Standard conformance for production use
Suppress: set suppress_production_warning=True

Why this matters: - Minimal has no governance contracts → uniform 300ms latency for all actions - Production apps need different latency for different risks (50ms chat vs 1000ms financial) - Without contracts, you can't negotiate performance/quality trade-offs

Target: <1 incident per quarter caused by Minimal tier misuse

Learn about production warnings

---

What's Missing in Minimal?

Optional features you can add later:

• Additional interventions - Nudge, Flag, Escalate, Halt (only OK/BLOCK in Minimal)
• Governance Contracts - Performance SLAs and latency budgets
• Trust debt - Earn autonomy through good behavior
• Advanced ARS/CTQ - Deep cognitive quality analysis
• Tripwires - Rate limits and safety bounds
• ReflectionDB - Persistent audit trail
• MCP/A2A integration - Framework connectors

All optional. Add them when you need them.

---

Next Steps

• Add More Features

  ---

  Incrementally add interventions, tripwires, and trust debt

  Standard Conformance

• Go to Production

  ---

  Upgrade to Standard with governance contracts and full features

  Standard Conformance

• Integrate Frameworks

  ---

  Auto-generate traces with LangChain, AutoGPT, or LlamaIndex

  Framework Integrations

• Learn Concepts

  ---

  Understand why ACGP exists and what problems it solves

  Why ACGP?