agdr-aki 1.8.3

Atomic Kernel Inference SDK for cryptographically-sealed, court-admissible AI decision records
Documentation

AKI + AgDR

Atomic Kernel Inference seals decisions at the moment of inference. Atomic Genesis Decision Record produces court-admissible proof. Together they create a stakeholder shield for regulated AI.

Why This Matters

AI decisions face regulatory scrutiny. Boards need defensible evidence when algorithms impact people. AKI and AgDR deliver cryptographic accountability that holds up in court.

What Is AKI?

AKI is the inference engine. It captures context, reasoning, and output at the exact moment a decision is made. Every capture is sealed with Ed25519 signatures and BLAKE3 hashes.

What Is AgDR?

AgDR is the decision record. It produces a tamper-evident, timestamped artifact that regulators can verify. Each record includes provenance, purpose, and cryptographic proof of integrity.

The Stakeholder Shield

Court-admissible records protect decision-makers. When AI systems face audit or litigation, AgDR provides the evidence trail. Cryptographic sealing means no one can alter the record after the fact.

Installation

Python Users

pip install agdr-aki

Rust Users

[dependencies] agdr-aki = "1.8"

Quick Start

Python Example

from agdr_aki import AKIEngine, PPPTriplet, HumanDeltaChain

engine = AKIEngine(":memory:") ppp = PPPTriplet("audit", "compliance", "regulatory_review") hdc = HumanDeltaChain("decision_123", True, "approved")

record = engine.capture( ctx={"user": "reviewer"}, prompt="Approve loan application?", reasoning_trace={"risk_score": 0.23, "policy_match": True}, output="approved", ppp_triplet=ppp, human_delta_chain=hdc )

print(f"Sealed: {record.hash}") print(f"Public key: {engine.public_key_hex()}")

Rust Example

use agdr_aki::{AKIEngine, monotonic_raw_nanos};

let mut engine = AKIEngine::new(":memory:".into()); let timestamp = monotonic_raw_nanos(); println!("Monotonic timestamp: {}", timestamp);

Technical Details

Cryptographic Guarantees

  • Ed25519 digital signatures for authenticity
  • BLAKE3 hashing for integrity
  • Merkle root chaining for sequence verification
  • CLOCK_MONOTONIC_RAW timestamps for tamper resistance

Compliance Alignment

  • PIPEDA: Audit trail for personal information handling
  • EU AI Act: Documentation for high-risk AI systems
  • Canada Evidence Act: Cryptographic proof admissible in court

Performance

Rust core achieved a 0.62 µs latency floor in testing. High-throughput systems scale linearly with core count. Prebuilt wheels require Python 3.12+. Source builds support 3.9+.

For Developers

Build from Source

Requires Rust toolchain + Python 3.8+ git clone https://github.com/aiccountability-source/AgDR-Phoenix cd AgDR-Phoenix python -m maturin develop # For Python testing cargo build --release # For Rust-only usage

Feature Flags

  • python (default): Enables PyO3 bindings
  • Build without Python: cargo build --no-default-features

License

Licensed under either of:

  • Apache License, Version 2.0
  • CC0 1.0 Universal

At your option. See LICENSE-APACHE and LICENSE-CC0 for details.

About Genesis Glass Foundation

Genesis Glass Foundation is a federally incorporated Canadian not-for-profit. We steward open standards for cryptographic accountability in AI systems. Our work supports ethical governance through verifiable decision records.

accountability.ai