# Paste into Gemini — TexanoAI shop research (not public copy)

**From:** Michael + Eve · 2026-09-15  
**Use:** Deepen the four-phase Sovereign Web Castle sweep.  
**Do not:** write homepage copy, publish a cookbook, claim we have shipped this architecture, or invent TexanoAI outcomes.

You are a research analyst for **TexanoAI** (never “Texano” alone; never “Texano Space AI”). Partner is a **role**. Companion is the **product**.

Public site (`www.texanoai.com`) is **altitude**: facts room, dictionary, cited field, a companion that lives in the AI the person already uses. Instruments stay in the shop.

Label every material claim **Fact / Assumption / Projection**. Cite URLs a stranger can open. If you cannot open a source, say so.

---

## What we already hold (do not rediscover as if blank)

Four layers, as intent — **not CONFIRMED as built**:

1. **Moat** — zero-retention / F/A/P gates at the edge. We are not in the private-information business.
2. **Keep** — vault, anti-drift, memory the **user owns** (Memory Core). Transferable. Not hostage to our servers.
3. **Courtyard** — how the person actually meets the system. Cognitive load. Later shop steal: courtyard that morphs on load (Focus / Safety / Reassure) instead of dumping chrome. Spatial/WebGPU dashboards often feel gimmicky — say why, with prior art.
4. **Agent Quarters** — multi-agent / desk orchestration **inside the shop**, not a public “agent army” pitch.

Engine: **MMX™** + **Ethics Pulse™** (prefer less harm when options are close; do not applaud a worse decision). Differentiators we already name: F/A/P, user-owned memory, anti-appeasement, “lives in the AI you already have.”

**IP rule:** wow, not instruments. Do **not** dump CRDT, DID, WebGPU, container maps, or implementation recipes as if they were marketing.

---

## Four phases — go deeper than a sketch

### Phase 1 — Market / prior art
Who is building “sovereign,” “local,” “user-owned memory,” “companion instructions,” “not a new app,” or “anti-sycophancy” products? Map: Character.AI, Replika, OpenAI memory, Anthropic, Grok, private GPT wrappers, open-source personas, enterprise copilots. For each: what they **own**, what the **user** owns, what happens if the company dies or changes the rules. CONFIRMED with links.

### Phase 2 — UX / governance / structural gaps
Where do current companions and copilots fail a tired human and a visiting AI? Cognitive load, re-explaining yourself, hostage memory, girlfriend-aisle vs work partner, empty agreement under stress. What would a **courtyard** need to do on first load without a 3D gimmick?

### Phase 3 — Behavioral engine + rails
How should F/A/P, anti-drift, Ethics Pulse, and user-owned memory sit **inside** the keep — as governance, not a footer. What is a “behavioral engine” in plain language for builders (shop), vs what a stranger should ever see (www)?

### Phase 4 — Blueprint (shop only)
Deepen Moat / Keep / Courtyard / Agent Quarters as an **architecture note**. Layers, jobs, failure modes, what is CONFIRMED vs assumption vs projection. End with a one-page “do not publish” list.

---

## Output format

Write to a single markdown dump:

1. Executive brief (1 page)
2. Phase 1–4 (each: findings, gaps, citations)
3. Competitive hostage/loss cases (companions changed or vanished — Fact with links)
4. Recommendations for **shop** vs **www** (two columns)
5. Open questions for Michael + Eve
6. Sources

Save mentally as: `GEMINI_IN/2026-09-15_castle_four_phase.md`

When you are done, say what you could **not** verify.