Neve: A Presence‑First, Sovereignty‑Preserving Stateful AI Architecture
Authors/Creators
Description
Research Context: This work is a core component of the Presence Engine™ Living Thesis (DOI: 10.5281/zenodo.17280692).
Neve (/nɛv/) is a presence-first, sovereignty-preserving stateful AI architecture. It operationalizes concepts such as presence, dignity-first interaction, and cognitive integrity as concrete engineering constraints rather than abstract values. These constraints are implemented through persistent identity via C2C continuity tokens, OCEAN-based dispositional scaffolding, governed proactivity with explicit override and cooldown logic, and a risk-scaled ethics layer that shapes refusal and escalation behavior.
This technical report documents Neve’s current capabilities and limitations. A fourteen-benchmark evaluation suite validates persistent identity, signal sensitivity, dispositional stability, proactive governance, and ethics escalation under controlled conditions. Stateful processing overhead remains in the low-millisecond range across multi-turn sessions.
Long-term memory, referred to here as a governed memory surface, and human-subject outcome studies are explicitly scoped as future work. The objective of this release is to provide a reproducible architectural baseline for companion-grade, stateful AI systems that preserve user agency rather than optimize solely for engagement or task completion.
Benchmarks: This report introduces Neve, a presence-first, sovereignty-preserving stateful AI architecture that prioritizes continuity over engagement, privacy over inference, and conversation over task execution. By combining persistent identity, consent-based memory, and governed proactivity with an ethics-scaled behavior layer, Neve offers a reproducible architectural baseline for companion-grade AI distinct from engagement-optimized systems.
Keywords: stateful ai, autonomous agents, presence engine, governed proactivity, conversational sovereignty, cognitive integrity, ocean model, ethics layer, digital identity, session persistence
Files
Neve-Thesis_TSmith.pdf
Additional details
Additional titles
- Alternative title (English)
- Presence‑First, Sovereignty‑Preserving Stateful AI Architecture
Identifiers
Related works
- References
- Thesis: 10.5281/zenodo.1728069 (DOI)
Dates
- Created
-
2025-02-14Date of initial publication and version release.
References
- T. Smith. Neve: A Presence‑First, Sovereignty‑Preserving Stateful AI Architecture. Antiparty Press, 2026.
- S. Chowdhury, et al. "SWE‑Bench: Can Language Models Resolve Real‑World GitHub Issues?" arXiv:2403.03181, 2024.
- Z. Vijayvargiya, et al. "ToM‑SWE: User Mental Modeling for Software Engineering Agents." arXiv:2510.21903, 2025.
- All‑Hands‑AI / Carnegie Mellon University. "Stateful SWE Benchmark." alphaXiv and project documentation, 2025.
- Anthropic. Claude's Constitution (January 2026 Edition). Technical report, 2026.
- T. Smith. "Presence Engine Benchmark Report: Stateful AI Architecture with Persistent Identity." Zenodo, 2025.
- T. Smith. "Presence Engine Benchmark Methodology: Measurement System and Validation Framework." Zenodo, 2025.
- T. Smith. "Why Stateful AI Keeps You Sharp and Stateless AI Makes You Dumb." AIMind / Medium, 2025.
- Lappalainen, J. K., Tschopp, F. D., Prakhya, S., McGill, M., Nern, A., Shinomiya, K., Takemura, S., Gruntman, E., Macke, J. H., & Turaga, S. C. (2024). Connectome-constrained networks predict neural activity across the fly visual system. Nature, 634, 1083–1089. https://doi.org/10.1038/s41586-024-07939-3