I build deterministic governance infrastructure for AI systems.

Phionyx treats large language model outputs as noisy cognitive measurements rather than final answers. The goal is to place a verifiable governance runtime between AI systems and end users: safety gates, ethics gates, telemetry, evaluation standards, state evolution, and audit-first control.

Latest (2026-05): Phionyx Core v0.4.0 is live on PyPI alongside three open-source MCP companion packages — phionyx-mcp-server (MCP trust boundary), phionyx-pipeline-mcp (agent self-claim gate), and phionyx-eval-inspect (Inspect AI bridge). They share one trace per session. See phionyx.ai/runtime-evidence for the argument and pip install phionyx-core==0.4.0 to try it.

• Phionyx Core SDK — deterministic AI governance runtime (v0.4.0 live on PyPI)
• Phionyx MCP Governance Stack — three-layer runtime evidence for AI coding agents (Claude Code, Cursor, Zed, JetBrains): outward MCP trust boundary (descriptor signing, RGE v0.2 envelope, audit chain) + inward self-claim gate (three-layer verification over the agent's own "fixed / tested / changed" declarations) + Inspect AI interoperability bridge (envelope chain → viewable .eval log); all share one trace per session
• HearthOS — bounded-authority household AI: the operating principle from Volume I of the Governance Trilogy, demonstrated end-to-end in three browser-only modules (Diagnostic, Weekly Reset, Boundary Script) backed by an open-source TypeScript reference implementation and a free printable Starter Kit
• Phionyx Evaluation Standard — behavioural reliability, safety, coherence, determinism, and long-term stability evaluation
• Governance Node Architecture — multi-gate AI control and release model
• Trace / Wheel & Balance — educational and narrative ecosystem for resilience, decision-making, and non-violent RPG-based learning (trace.phionyx.ai · @trace_phionyx)

• LLM output is not truth; it is a signal requiring governance.
• AI systems need runtime control, not only prompt-level safety.
• Safety, coherence, and telemetry should be structured before response release.
• Evaluation must include behavioural stability, not only benchmark performance.
• Human-facing AI should be explainable, auditable, and interruptible.

• phionyx-research — runtime evidence layer for agentic AI (Python; PyPI: phionyx-core)
• phionyx-mcp-server — MCP trust boundary: descriptor signing, signed envelopes, audit chain over third-party MCP tool calls (aligned with arXiv:2512.06556 threat taxonomy)
• phionyx-pipeline-mcp — agent self-claim gate: verifies what the agent says it did against the repository's actual diff
• phionyx-eval-inspect — Inspect AI bridge: Phionyx runtime evidence exported into Inspect AI .eval evaluation logs (interop-only, no AISI endorsement claim)
• phionyx-evaluation-standard — vendor-independent evaluation standard for agentic AI runtimes
• hearthos — bounded-authority household AI orchestration; TypeScript reference implementation, browser-only demo, Starter Kit PDF (AGPL-3.0)

• A model saying "fixed" is not evidence (2026-05-22 · X Article) — https://x.com/phionyx_ai/status/2057860001117454685
• Persistent Worlds Need Deterministic Governance (2026-05-22 · Substack post 5) — https://phionyxresearch.substack.com/p/persistent-worlds-need-deterministic
• MCP Connects Tools. Runtime Evidence Keeps Agents Accountable. (2026-05-19 · X Article) — https://x.com/phionyx_ai/status/2056811861782274094

• Website: https://phionyx.ai · runtime evidence campaign page
• HearthOS demo: https://phionyx.ai/hearthos
• Substack: https://phionyxresearch.substack.com
• X: https://x.com/phionyx_ai

If runtime evidence for agentic AI is a problem you have, watch phionyx-research to get email updates when we ship new experiments.