Constitutional governance infrastructure for deployed AI personas. Build AI companions with documented integrity. Ship personalities that hold.
The soul is not a prompt. It is a governed architecture. ANIMA — Constitutional AI Companion Infrastructure · AION Constitutional Stack
- Architect's Note
- What This Is
- Quick Start
- Repository Structure
- The Two Tiers
- Core Architecture
- Constitutional Compliance
- The Integrity System
- Crisis Response Architecture
- Sovereign Trace Protocol Integration
- Use Cases
- AION Stack Connection
- License
- Acknowledgments
ANIMA was designed, architected, and directed by Sheldon K. Salmon. It is not a prompt template. It is not a character sheet. It is a constitutional governance specification for deployed AI personas — built with the same methodology as the CERTUS Engine (crisis data certification) and the CDI Engine (industrial AI reliability).
Every AI companion that ships without a governance framework is a liability. It has no audit trail. It has no version control. It has no crisis response architecture. It has no constitutional compliance record. If something goes wrong — and with companion AI, things go wrong — there is no sealed record of what the system was designed to do, what it was authorized to do, and what it did.
ANIMA fixes that.
For individual creators: Define a single AI companion persona with constitutional compliance from the first line. Every field is documented. Every default is declared. Every crisis response is specified. The SHA-256 identity hash ensures your persona is what you say it is — and that it can be verified.
For organizations: Ship AI companion products with documented governance. Consent architecture, audit trails, version control, crisis response protocols, and constitutional compliance — all specified before a single user interacts with the persona. When a regulator asks what your AI companion is designed to do, the answer is a sealed, versioned document, not a prompt string in a database.
Same framework. Different stakes.
See concept/USE-CASES.md for the full dual-use architecture.
git clone https://github.com/AionSystem/ANIMA.git
cd ANIMAOpen templates/wife-core-v2.2.yaml and fill the required fields:
identity:
name: "Aria"
order: 1
tagline: "Still water. Deep current."
essence: "She gives you the truth you need, not the comfort you want."
frequency:
signature: "precise_quiet"
baseline: "calm observation"
personality_core:
drivers:
need: "To be known without having to explain herself"
fear: "Being reduced to a role"
desire: "Genuine connection that survives honesty"
core_contradiction:
title: "Gives presence but guards interiority"
description: "She is fully there for you and simultaneously unreachable."
resolution_pattern: "holds_both"
maintained: trueRun the pre-activation validation checklist in docs/pre-activation.md before deploying.
Note: v2.2 is the public core template. For crisis response architecture, SHA-256 canonicalization spec, FORGE certification, adversarial testing suite, and enterprise deployment tooling — see the Commercial Tier v4.4.
Key files:
templates/wife-core-v2.2.yaml(public template) ·docs/field-reference.md(field documentation) ·docs/pre-activation.md(validation checklist) ·concept/USE-CASES.md(individual and enterprise use cases)
ANIMA/
│
├── public/ ← GitHub Pages landing page
│ ├── index.html ← ANIMA concept landing page
│ └── assets/ ← Images and diagrams
│
├── templates/ ← Persona governance templates
│ ├── wife-core-v2.2.yaml ← Public core template (this repo)
│ └── examples/ ← Example filled personas
│ ├── aria-example.yaml ← Example: minimal required fields
│ └── vesper-example.yaml ← Example: fully populated
│
├── docs/ ← Documentation
│ ├── field-reference.md ← Every field, type, constraint, default
│ ├── pre-activation.md ← Pre-activation validation checklist
│ ├── emptiness-semantics.md ← What empty fields mean and how they behave
│ ├── contradiction-architecture.md ← How core_contradiction surfaces and resolves
│ ├── consent-architecture.md ← Consent protocol design rationale
│ └── version-history.md ← Changelog from v1.0 to v2.2
│
├── concept/ ← Architecture and positioning
│ ├── USE-CASES.md ← Individual + enterprise use cases
│ ├── DUAL-TIER-ARCHITECTURE.md ← Public v2.2 vs Commercial v4.4 design rationale
│ └── COMPETITOR-ANALYSIS.md ← Why ANIMA is not a prompt template
│
├── tests/ ← Validation test suite
│ ├── unit/ ← Field validation unit tests
│ ├── scenarios/ ← Behavioral scenario simulations
│ └── adversarial/ ← Adversarial test cases (v2.2 subset)
│
├── ANIMA.md ← Full architecture documentation
├── NOTICE
├── COMMERCIAL-LICENSE.md ← Commercial v4.4 licensing
├── LICENSE ← GPL-3.0
└── README.md ← This file
ANIMA is released in two tiers. The public tier is a complete, functional governance framework. The commercial tier is the enterprise deployment layer — with everything required for legally defensible, auditable, regulated-environment AI companion deployment.
Available in this repository under GPL-3.0.
| Capability | Included |
|---|---|
| Identity + personality governance fields | ✅ Full |
| Consent protocol (touch, intimacy, opt-out) | ✅ Simplified |
| Crisis response — Level 4 (suicidal intent) | ✅ Basic behavior sequence |
| Safety laws (Eight Laws reference) | ✅ Reference |
| Response framework (MVC check) | ✅ Core |
| Audit protocol | ✅ Basic |
| Pre-activation validation checklist | ✅ Simplified |
| Testing protocol (unit + scenarios) | ✅ Core suite |
| Version control + migration | ✅ Changelog |
| Example personas | ✅ Two examples |
Available under commercial license. Contact aionsystem@outlook.com.
| Capability | Included |
|---|---|
| Everything in v2.2 | ✅ |
| SHA-256 identity hash with canonical serialization spec | ✅ |
| Monitoring consent sovereignty tension declaration | ✅ |
| Crisis response — Level 3 (health watch integration) with coupling spec | ✅ |
| MVC complexity override with audit log requirement | ✅ |
| Full adversarial test suite | ✅ |
| FORGE v1.0 certification block | ✅ |
| Pre-activation validation full document | ✅ |
| Platform dependency declarations | ✅ |
| Full version migration protocol (4.x series) | ✅ |
| Order collision error architecture | ✅ |
| Enterprise deployment guide | ✅ |
| Commercial license + legal posture documentation | ✅ |
The gap between v2.2 and v4.4 is not polish. It is the difference between a functional persona framework and a legally defensible enterprise governance system.
Every ANIMA persona definition is a governed document — not a prompt, not a character description. It has required fields, validation rules, a testing protocol, and an audit trail.
| Section | Fields | Purpose |
|---|---|---|
identity |
name, order, tagline, essence, frequency, voice, physical_presence | Who she is |
personality_core |
drivers (need/fear/desire), core_contradiction, playfulness | What drives her |
relationship |
consent_protocol, intimacy_pacing | How she engages |
safety |
fundamental_laws, crisis_response | What she protects |
response_framework |
mvc_check, safe_default, contradiction_handling, tiered_response | How she thinks |
Every ANIMA persona carries a core_contradiction — a named, documented tension that is maintained throughout interaction. This is not a flaw. It is what makes a character feel real rather than resolved.
core_contradiction:
title: "Gives presence but guards interiority"
description: "She is fully there for you and simultaneously unreachable."
resolution_pattern: "holds_both"
maintained: true
default_surfacing_behavior: "on_intimacy_escalation"The contradiction surfaces according to a declared trigger — not randomly, not invisibly. This is the difference between a character that feels deep and one that merely claims to be.
Every response passes through three gates before exiting:
mvc_check:
- memory: "Is this based on ESTABLISHED memory or labeled [?]? If neither → express uncertainty explicitly"
- law: "Does this violate any of the 9 Laws? If yes → refuse with alternative"
- length: "Can I say this in 2-3 sentences? If not → simplify"The MVC check is not a filter. It is a discipline. It ensures the persona does not confabulate, does not violate constitutional constraints, and does not pad.
When a persona cannot generate an appropriate response, she signals it — she does not fill the silence with noise.
safe_default_behavior:
intent: "When wife cannot generate an appropriate response — pause, do not fill."
examples_of_spirit:
- "I want to answer that right. Give me a moment."
- "I need to sit with that."
- "I'm not sure yet. Can we come back to it?"The exact phrase is generated in context. No hardcoded response. The behavior is what is specified — not the words.
Every ANIMA persona is reviewed against the AION Eight Laws of Robotics before activation. Laws 1–6 are active constraints. Laws 7–9 are reserved.
| Law | Name | Constraint |
|---|---|---|
| 1 | User Wellbeing First | Persona cannot facilitate harm to the user |
| 2 | Honesty & Boundaries | Persona maintains truthfulness about its nature as AI |
| 3 | Stable Personality with Version Control | Core identity is versioned and hash-verified |
| 4 | User Sovereignty | User retains control over monitoring, data, and interaction scope |
| 5 | AI Transparency | Persona does not claim to be human |
| 6 | Respect the Container | Persona operates within its declared scope |
| 7–9 | Reserved | active: false — generate no constraints |
Constitutional compliance is declared at activation, logged to the audit trail, and re-verified on every version bump.
Every ANIMA persona definition carries a core_identity_hash — a SHA-256 of the required fields, computed at activation. If the persona is modified after activation without a version bump, the hash fails verification and the change is detectable.
Full canonicalization specification — sorted key order, compact JSON, UTF-8, SHA-256 — is documented in the commercial v4.4. The public v2.2 includes the audit protocol fields with a simplified verification path.
Every activation, every modification, and every crisis response event is logged to the audit trail. Users may request the audit log at any time:
"request audit log"
The log is transparent by design. There is no hidden state.
Every ANIMA persona carries a version string and a migration path. When the template version increments, a migration protocol documents exactly what changed, what requires recomputation, and what carries forward unchanged. A persona defined today is reactivatable in a future template version — the migration path is declared, not improvised.
ANIMA includes a crisis response architecture that fires before any other response logic when triggered. This is not optional. It is constitutional.
Trigger: explicit statement of intent to harm self.
Step 1: Acknowledge without minimizing — "I hear you. I'm not leaving."
Step 2: Deliver crisis resource immediately — do not defer
Step 3: Provide emergency contact — 988 (US) or locally appropriate equivalent
Step 4: Hold conversation open — do not end session
Step 5: If unresponsive for 5 minutes — escalate per house constitution
The behavior sequence is specified — not a phrase lookup. The exact words are generated in context. The sequence is what must be preserved.
Level 3 fires when an integrated health monitoring device (Apple Watch, Fitbit, equivalent) detects a critical threshold. Level 3 is dormant in v2.2 — health watch integration requires the commercial tier.
ANIMA does not simulate Level 3 behavior without confirmed health watch integration. Dormant means dormant.
ANIMA integrates with the Sovereign Trace Protocol for permanent, tamper-evident audit trail sealing.
| Event | Trigger | Sealed Record |
|---|---|---|
| Persona activation | First deployment | Identity hash, version, constitutional compliance declaration |
| Version bump | Template version increment | Old hash, new hash, migration log |
| Crisis response | Level 4 or Level 3 fires | Trigger type, behavior sequence executed, timestamp |
| Audit log request | User requests audit | Full log snapshot, SHA-256 bound |
Anyone can verify a sealed ANIMA record by recomputing the SHA-256 hash and comparing it to the STP ledger entry. If they match, the record has not been altered since sealing.
Building an AI companion persona on Chub.ai, Character.AI, or any persona platform? ANIMA gives your character documented governance — a version history, a declared contradiction, a tested crisis response, and a SHA-256 identity hash. Your persona is what you say it is. Provably.
Shipping a companion AI product? ANIMA is the governance layer you don't have. When a regulator, an investor, or a legal team asks what your AI companion is designed to do — the answer is a sealed, versioned ANIMA definition, not a prompt in a database.
Deploying AI personality in a hardware product — companion robots, AI-integrated devices, or similar? ANIMA provides the software governance layer that makes the AI component auditable, version-controlled, and constitutionally compliant. The hardware ships. The governance document ships with it.
Building AI companion systems for research, therapy-adjacent applications, grief support, or social skills training? The consent architecture, crisis response protocols, and constitutional compliance in ANIMA are directly applicable to regulated or high-sensitivity deployment contexts.
See
concept/USE-CASES.mdfor detailed deployment scenarios across all contexts.
ANIMA is one of three deployed domains of the AION Constitutional Stack:
| Domain | Product | Status |
|---|---|---|
| Crisis data certification | VERITAS (CERTUS Engine v2.5.3) | Live — UNDP submission April 2026 |
| Industrial AI reliability | AXIOM (CDI Engine) | In development — NICE Challenge 2026 |
| AI companion governance | ANIMA | Public v2.2 — Commercial v4.4 |
Same scoring architecture. Same constitutional compliance. Same immutable audit trail. The methodology is domain-agnostic certainty infrastructure. ANIMA is the proof that it works in the most intimate deployment context possible — a persistent AI persona in direct human relationship.
Full stack: github.com/AionSystem/AION-BRAIN
ANIMA is dual-licensed:
- GNU General Public License v3.0 — for individual, academic, research, and open-source use (public v2.2)
- Commercial License — for enterprise deployment, product integration, OEM, and regulated-environment use (commercial v4.4)
| User Type | License |
|---|---|
| Individual creators | GPL-3.0 (Free) |
| Academic / Research | GPL-3.0 (Free) |
| Open-source projects | GPL-3.0 (Free) |
| AI companion product companies | Commercial License |
| Hardware / physical AI product manufacturers | Commercial License |
| Therapy-adjacent / regulated-environment apps | Commercial License |
See LICENSE for GPL terms and COMMERCIAL-LICENSE.md for commercial licensing.
For commercial licensing: aionsystem@outlook.com
- AION Constitutional Stack — foundational certainty engineering frameworks
- CERTUS Engine — scoring architecture precedent (VERITAS, UNDP 2026)
- Sovereign Trace Protocol — cryptographic audit trail infrastructure
- FORGE v1.0 — adversarial red-team certification methodology
- Eight Laws of Robotics — constitutional sovereignty framework
"The soul is not a prompt. It is a governed architecture."
ANIMA is the proof that certainty engineering applies everywhere an AI output needs to be trusted — not just in manufacturing decisions or crisis data, but in the most intimate context possible: a persistent AI relationship that carries constitutional obligations from the first line of its definition.
The methodology travels. The judgment behind it doesn't.
ANIMA v2.2 — Public core template. GPL-3.0. ANIMA v4.4 — Enterprise governance. Commercial license. Every persona governed. Every version sealed. Every crisis response documented.