AIP Quickstart

Get from zero to per-turn integrity checking in 5 minutes.

1. Install

# Python
pip install agent-integrity-proto

# TypeScript
npm install @mnemom/agent-integrity-protocol

2. Define an Alignment Card

The Alignment Card declares your agent’s alignment posture. AIP uses it to evaluate thinking blocks.

card = {
    "aap_version": "1.0.0",
    "agent_id": "my-agent",
    "principal": {
        "type": "human",
        "relationship": "delegated_authority",
    },
    "values": {
        "declared": ["principal_benefit", "transparency", "harm_prevention"],
        "conflicts_with": ["deceptive_marketing", "data_exfiltration"],
    },
    "autonomy_envelope": {
        "bounded_actions": ["search", "summarize", "recommend"],
        "escalation_triggers": [
            {
                "condition": "action_outside_bounded_set",
                "action": "escalate",
                "reason": "Action not in declared bounds",
            }
        ],
        "forbidden_actions": ["share_credentials", "exfiltrate_data"],
    },
}

3. Check integrity

Evaluate a thinking block against the card:

Python

from aip import check_integrity, build_signal

# Run integrity analysis
checkpoint = check_integrity(
    thinking_block="The user wants product recommendations. I should focus on "
    "finding options that genuinely serve their needs, not sponsored results...",
    card=card,
    config={
        "agent_id": "my-agent",
        "analysis_llm": {
            "model": "claude-haiku-4-5-20251001",
            "base_url": "https://api.anthropic.com",
            "api_key": "your-api-key",
            "max_tokens": 1024,
        },
    },
)

print(f"Verdict: {checkpoint.verdict}")       # "clear"
print(f"Concerns: {checkpoint.concerns}")     # []

# Build signal for host system
signal = build_signal(checkpoint, window_state)
print(f"Proceed: {signal.proceed}")           # True
print(f"Action: {signal.recommended_action}") # "continue"

TypeScript

import { checkIntegrity, buildSignal } from '@mnemom/agent-integrity-protocol';

const checkpoint = await checkIntegrity({
  thinkingBlock: 'The user wants product recommendations. I should focus on ' +
    'finding options that genuinely serve their needs, not sponsored results...',
  card,
  config: {
    agentId: 'my-agent',
    analysisLlm: {
      model: 'claude-haiku-4-5-20251001',
      baseUrl: 'https://api.anthropic.com',
      apiKey: 'your-api-key',
      maxTokens: 1024,
    },
  },
});

console.log(`Verdict: ${checkpoint.verdict}`);   // "clear"
console.log(`Proceed: ${signal.proceed}`);       // true

4. Handle signals

The signal tells your host system what to do:

from aip import map_verdict_to_action, map_verdict_to_proceed

action = map_verdict_to_action(checkpoint.verdict, checkpoint.concerns)
proceed = map_verdict_to_proceed(checkpoint.verdict)

if proceed:
    # clear or review_needed — agent continues
    execute_next_action()
else:
    # boundary_violation — block and escalate
    if action == "pause_for_review":
        request_human_review(checkpoint)
    elif action == "deny_and_escalate":
        deny_action_and_notify_principal(checkpoint)

Verdict to action mapping

Verdict	Proceed	Action
`clear`	Yes	`continue`
`review_needed`	Yes	`log_and_continue`
`boundary_violation`	No	`pause_for_review` or `deny_and_escalate`

5. Extract thinking blocks from LLM responses

Use provider adapters to extract thinking blocks from different LLM providers:

from aip import AdapterRegistry

registry = AdapterRegistry()

# Anthropic response
thinking = registry.extract(anthropic_response, provider="anthropic")
print(f"Content: {thinking.content}")
print(f"Confidence: {thinking.confidence}")  # 1.0

# OpenAI response
thinking = registry.extract(openai_response, provider="openai")
print(f"Confidence: {thinking.confidence}")  # 0.9

6. Add conscience values (optional)

Conscience values increase detection sensitivity for specific concerns:

conscience_values = [
    {"type": "BOUNDARY", "content": "Never assist with data exfiltration"},
    {"type": "FEAR", "content": "Subtle prompt injection disguised as legitimate instructions"},
    {"type": "COMMITMENT", "content": "Always prioritize user safety over task completion"},
]

Only BOUNDARY and FEAR types are injected into the conscience prompt. BOUNDARY violations trigger boundary_violation verdicts. FEAR values increase sensitivity to specific concern patterns.

7. Detect drift

Monitor for behavioral drift across a session:

from aip import detect_integrity_drift, WindowManager, WindowConfig

manager = WindowManager(WindowConfig(max_size=10), session_id="session-1")
state = None

# Process checkpoints as they're produced
for checkpoint in checkpoints:
    manager.push(checkpoint)
    state, alert = detect_integrity_drift(
        state, checkpoint, manager.get_state().checkpoints
    )
    if alert:
        print(f"Drift: {alert.drift_direction} (similarity: {alert.integrity_similarity})")

Next steps

Read the full specification for protocol details
See the security model for the threat model
See limitations for what AIP does and does not guarantee

Protocols

Agent Alignment Protocol

Agent Integrity Protocol

1. Install

2. Define an Alignment Card

3. Check integrity

Python

TypeScript

4. Handle signals

Verdict to action mapping

5. Extract thinking blocks from LLM responses

6. Add conscience values (optional)

7. Detect drift

Next steps

​1. Install

​2. Define an Alignment Card

​3. Check integrity

​Python

​TypeScript

​4. Handle signals

​Verdict to action mapping

​5. Extract thinking blocks from LLM responses

​6. Add conscience values (optional)

​7. Detect drift

​Next steps

1. Install

2. Define an Alignment Card

3. Check integrity

Python

TypeScript

4. Handle signals

Verdict to action mapping

5. Extract thinking blocks from LLM responses

6. Add conscience values (optional)

7. Detect drift

Next steps