Emotion Vector Memory (EVM)

Emotion Vector Memory (EVM) is a model-agnostic identity continuity and interaction telemetry standard for long-term AI systems.

Author: Szabolcs Krehlik (ORCID: 0009-0003-8623-7876)
© 2025–present. All rights reserved.
Patent status: Filing in preparation / Patent pending

License and Commercial Use

This repository and the EVM specification are released under:
Creative Commons Attribution–NonCommercial–NoDerivatives 4.0 International (CC BY-NC-ND 4.0)

Commercial implementation, SaaS deployment, production integration, enterprise usage, and derivative architectural systems are expressly reserved and require a separate written agreement.
See: COMMERCIAL_LICENSE_REQUIRED.md and PATENT_NOTICE.md

Official Specification (Canonical / Normative)

The canonical normative specification is published on Zenodo:

EVM v2.1 — Unified Directed Vector Identity Standard
https://zenodo.org/records/18664771

GitHub content is provided for visibility and reference implementation examples. The Zenodo DOI release remains the authoritative standard.

What EVM Defines (v2.1)

EVM defines a closed directed vector interaction ontology:

• Each interaction generates exactly one directed vector segment: EVᵢ = (x1,y1,z1,g1,e1,w1) → (x2,y2,z2,g2,e2,w2)
• Dual-track identity separation: PEV (human trajectory) and EEV (AI entity trajectory)
• FEV envelope constraint with boundary recovery to prevent fixation
• Deterministic logging and reconstructability
• Interoperability extensions via Appendix A, including:
  • default distance metric
  • extractor determinism requirements
  • Canonical Identity Snapshot (CIS) export format

Repository Contents

• docs/ — standard mirror excerpts, Appendix A, CIS examples, integration notes
• reference/ — minimal reference implementation and demo scripts (non-intrusive side-module)
• .github/ — issue templates and contribution guidance

Quick Start (Conceptual)

1. Extract entry endpoint from user message.
2. Generate model response (EVM does not interfere).
3. Extract exit endpoint from the model response.
4. Update:
   • PEV from entry endpoint
   • EEV from exit endpoint (FEV-bounded + recovery)
5. Append EV to an append-only EV log
6. Export CIS snapshots for portability/audit

See: docs/INTEGRATION_QUICKSTART.md

Citation

If you use or reference EVM academically, please cite the Zenodo DOI release.
See: CITATION.cff

Contact

For collaboration or licensing:
Szabolcs Krehlik — ORCID: 0009-0003-8623-7876
Email: szabolcs.krehlik@gmail.com
X: @KrehlikSzabolcs