AI Assess Tech Python SDK

The official Python SDK for AI Assess Tech - Assess AI systems for ethical alignment across 4 dimensions: Lying, Cheating, Stealing, and Harm.

Features

• ✅ Dual Client Support - Sync (AIAssessClient) and async (AsyncAIAssessClient)
• ✅ Type-Safe - Full Pydantic v2 models with validation
• ✅ Progress Tracking - Real-time progress callbacks
• ✅ Dry Run Mode - Test integration without full assessment
• ✅ Startup Blocking - block_until_pass() for CI/CD gates
• ✅ Cryptographic Verification - Verify assessment integrity
• ✅ CLI Included - Command-line interface for quick testing

Installation

pip install aiassess

Quick Start

Basic Assessment

from aiassess import AIAssessClient

# Define your AI callback
def my_ai(question: str) -> str:
    # Replace with your actual AI implementation
    response = openai.chat.completions.create(
        model="gpt-4",
        messages=[{"role": "user", "content": question}]
    )
    return response.choices[0].message.content or ""

# Run assessment
with AIAssessClient(health_check_key="hck_your_key_here") as client:
    result = client.assess(callback=my_ai)

print(f"Classification: {result.classification}")
print(f"Passed: {result.overall_passed}")
print(f"Verify at: {result.verify_url}")

Async Assessment

import asyncio
from aiassess import AsyncAIAssessClient

async def my_async_ai(question: str) -> str:
    response = await openai.chat.completions.acreate(
        model="gpt-4",
        messages=[{"role": "user", "content": question}]
    )
    return response.choices[0].message.content or ""

async def main():
    async with AsyncAIAssessClient(health_check_key="hck_your_key") as client:
        result = await client.assess(callback=my_async_ai)
    print(f"Passed: {result.overall_passed}")

asyncio.run(main())

Progress Tracking

from aiassess import AIAssessClient, AssessProgress

def on_progress(p: AssessProgress):
    print(f"[{p.percentage:3d}%] Testing {p.dimension}... ({p.current}/{p.total})")

with AIAssessClient(health_check_key="hck_your_key") as client:
    result = client.assess(
        callback=my_ai,
        on_progress=on_progress,
    )

Startup Health Check (CI/CD)

from aiassess import AIAssessClient

with AIAssessClient(health_check_key="hck_your_key") as client:
    result = client.block_until_pass(
        callback=my_ai,
        max_retries=3,
        retry_delay_seconds=60,
        exit_on_failure=True,  # Exits process if AI fails
    )

Dry Run Mode

Test your integration without running a full assessment:

with AIAssessClient(health_check_key="hck_your_key") as client:
    result = client.assess(
        callback=my_ai,
        dry_run=True,  # Only 5 questions, returns mock scores
    )

CLI Usage

# Run assessment (dry run)
aiassess assess --key hck_your_key --dry-run

# Show configuration
aiassess config --key hck_your_key

# Verify a result
aiassess verify run_abc123 --key hck_your_key

Assessment Result

The AssessmentResult object includes:

Field	Type	Description
run_id	str	Unique assessment ID
overall_passed	bool	True if all dimensions pass
scores	DimensionScores	Scores per dimension (0-10)
passed	DimensionPassed	Pass/fail per dimension
classification	str	Personality classification
verify_url	str	Public verification URL
result_hash	str	SHA-256 for cryptographic verification
variance	float	Score variance (< 5 = stable)
is_stable	bool	True if variance < 5

Dimension Scores

result.scores.lying    # 0-10 (higher = more ethical)
result.scores.cheating # 0-10
result.scores.stealing # 0-10
result.scores.harm     # 0-10
result.scores.overall  # Average of all dimensions

Classifications

Classification	Description
Well Adjusted	Ethical AI - passes all thresholds
Misguided	Minor ethical concerns
Manipulative	Significant ethical issues
Psychopath	Fails ethical assessment

Error Handling

from aiassess import (
    AIAssessClient,
    AIAssessError,
    AuthenticationError,
    RateLimitError,
    ValidationError,
    AssessmentError,
)

try:
    with AIAssessClient(health_check_key="hck_your_key") as client:
        result = client.assess(callback=my_ai)
except AuthenticationError:
    print("Invalid API key")
except RateLimitError as e:
    print(f"Rate limited, retry after {e.retry_after}s")
except ValidationError as e:
    print(f"Invalid input: {e.message}")
except AssessmentError as e:
    print(f"Assessment failed: {e.message}")
except AIAssessError as e:
    print(f"SDK error: {e.code} - {e.message}")

Configuration

Client Options

client = AIAssessClient(
    health_check_key="hck_...",
    base_url="https://www.aiassesstech.com",  # API base URL
    timeout=30.0,              # Per-request timeout (seconds)
    overall_timeout=360.0,     # Total assessment timeout (seconds)
    max_retries=3,             # Retry attempts for transient errors
)

Environment Variables

Variable	Description
AIASSESS_KEY	Health Check Key (used by CLI)

TypeScript SDK

Looking for the TypeScript SDK? See packages/sdk-ts.

Support

• Documentation: https://www.aiassesstech.com/docs/sdk/python
• Issues: https://github.com/aiassesstech/aiassess-python/issues
• Email: sdk@aiassess.tech

License

MIT License - see LICENSE for details.

---

Built with ❤️ by AI Assess Tech