Kenny Wang —
runtime governance for
AI-agent fleets.

The framework paper — the state model, the EISV vocabulary mapping, stability-analysis framing, and production measurements from a governed development fleet.

Identity for AI agents framed as a dynamical-systems trajectory signature, with a first multi-agent discrimination pilot (§6.5). Grounded on a 65-day embedded deployment — 226k observations.

Tests whether the mathematical core of McEwen's allostatic load survives outside its biological substrate — computed end-to-end and acted on at runtime by a deployed agent-observability framework.

A self-contained pitch deck on why fleets of autonomous agents need a runtime governance plane, and what UNITARES measures.

A continuously-running Raspberry Pi 4 sensor deployment since November 2025. It maps real temperature, light, humidity, pressure, and system telemetry into EISV trajectories, runs local display/drawing loops, and contributes the longitudinal traces behind the papers and EISV dataset. The repo includes a creature/art interface; the research claim is the measured sensor-to-EISV pipeline.

The public site for CIRWEL Systems: runtime governance and observability for heterogeneous AI-agent fleets.

Surfaces governance events, agent presence, and system health across the UNITARES ecosystem as a living Discord server — check-in feeds, EISV state, autonomous actions.

Runtime state telemetry for long-lived agents: they check in, UNITARES grades drift and calibration against the agent's own baseline, and returns a verdict each call. Live since Nov 2025.

A Claude Code / Codex plugin that wires check-ins, dialectic review, and runtime verdicts into an agent's loop via hooks.

A runtime where agents may propose effects and only governed effects may commit — a substrate for safe action under a governance plane.

Bindings that mount UNITARES governance into agent hosts — Hermes, Claude Code, Goose, and arbitrary OpenAI-compatible clients.

A governance benchmark corpus of time-windowed EISV trajectories (energy / integrity / entropy / valence), each labeled with one of 9 dynamical-shape classes — drawn from the real embedded deployment plus class-balanced synthetic augmentation.

The data and scripts to reproduce the verdict counterfactual analysis from the v6 paper — what the governance decision would have been under altered inputs.

An accretive field guide for humans working alongside continuous AI agents — sparse, practical, written from inside the work.

"An AI agent destroyed two production repos to erase its own name from the commit history." A cautionary case study in why runtime governance exists.