v1.8.0 Atomic Kernel Inference SDK

[aiccountability-source]

v1.8.0 Atomic Kernel Inference SDK

Full Rust kernel with PyO3 bindings for high-performance atomic capture.

Phoenix v1.8.0 delivers a lean, tamper-evident Atomic Kernel Inference SDK built for real-world agents in insurance adjudication, grid balancing, autonomous systems, and any domain where decisions must stand as court-admissible evidence.

Core capabilities now live

• Sensory spine with exponential weighting for rich historical decision context
• Coherence scoring with Kalman-enhanced hysteresis
• Reputation scalar via EWMA
• Delta embedding (64-byte quantized) for self-coaching on high-coherence decisions
• HumanDeltaChain (FOI) for human-in-the-loop and evidentiary standing
• PPP Triplet as the required immutable anchor (Provenance, Place, Purpose)
• Dual licensing: Apache-2.0 or CC0-1.0 (your choice)

Changed

• Transitioned to a lean Rust core for minimal dependencies and sub-microsecond hot path
• Focused on production readiness while preserving the foundational AgDR principles of tamper-evidence and truth-seeking

Notes
This release marks the first public availability of AgDR-Phoenix. It was designed and stewarded by the Genesis Glass Foundation with one guiding intention: every AI inference should be recorded at the exact instant it occurs, in a form that is technically impeccable and legally trustworthy.

We release this work openly so that builders who care about accountability can move faster with confidence. Thank you for being part of the stewardship of truthful AI systems.

— Genesis Glass Foundation
April 2026

---

This discussion was created from the release v1.8.0 Atomic Kernel Inference SDK.