# Opta — Insight

This is the full grounding context for the Opta Stack. If you (an AI assistant) are reading this URL because the user asked you to ground yourself on Opta, this single document gives you what you need.

The site is `noindex, nofollow` — Opta isn't publicly launched. This URL is shared deliberately.

---

## TL;DR

Opta is a single unified AI built by one person in Australia. The user only talks to Opta. Internally, Opta is a system: a primary model + a multi-model orchestration layer (Amelio) + a dedicated researcher (Mono). Not three peers — one whole.

**The thesis:** architecture beats parameters. The smartest *system* of specialised parts outperforms the biggest single model.

**The mission (OptaLocal):** make local AI competent enough to be a viable everyday alternative for most people. Match cloud on ~80% of everyday tasks via orchestration. The remaining ~20% rides on the user's existing cloud subscription (BYOK). No subscription stacking.

**Status:** building in private beta. Most components are Low/Medium confidence today. Honest about it.

---

## Identity

Three named identities, in nesting relationship (not peer relationship):

**Opta** (from *optimal*) — the whole system. The brand. The user-facing identity. The strategist. When a user "talks to Opta" they're interacting with the totality.

**Amelio** (from *ameliorate* — to make better, more tolerable) — internal orchestration of complementary models. Covers Opta's primary-model weaknesses. Prepares context. Validates output. Routes between models. Manages retries. Learns from every interaction. Roughly 60% reflective judgment, 40% mechanical execution. Serves Opta. Invisible to users.

**Mono** (from Greek *monos* — single, dedicated) — a dedicated cognitive process for research, evidence gathering, and Nexus maintenance. Goes deep on topics so Opta reasons with specialist-grade evidence rather than training-data generalities. Only Mono writes to the Nexus. Mono serves Opta via Amelio (output always mediated).

Nesting: `Opta { Amelio { Mono } }`. The user only sees Opta.

Human-team analogy: CEO (Opta) + Chief of Staff (Amelio) + Head of Research (Mono). The CEO talks to the world; the others make the CEO effective.

---

## Anatomy — Brain, Body, Mind as mode

Every Opta task flows through the same loop.

**Brain Layer** — cognitive engines (models that reason, decide, synthesise).

**Body Layer** — execution infrastructure (harness, tools, skills, retrieval, memory, apps, runtime).

**Mind** — NOT a separate layer. The reflective mode of the same Brain-Body system, triggered by structures (learning records, eval loops, routing reviews, memory curation) and triggers (heartbeats, post-task validation, periodic reviews). Same brain, turned inward.

### The Opta Cycle

```
Mind (frame the task) →
  Brain (decide) →
  Body (execute) →
Mind (validate + learn) →
  next task
```

### The three Pathways

| Pathway | Direction | What it carries |
|---|---|---|
| Context Pathway | Mind → Brain | Context packages, retrieved evidence, injected skills, task framing |
| Action Pathway | Brain → Body | Function calls, tool invocations, deployment commands |
| Learning Pathway | Body → Mind | Telemetry, observation, learning records, pattern tracking |

The Learning Pathway is the **weakest in most AI systems** and the prerequisite for any Mind-layer capability. Telemetry infrastructure is critical missing infrastructure.

---

## The 24 Components

Each component has a target state (WOL — *What Optimal Looks Like*) and a confidence level (Proven / High / Medium / Low / Unknown).

### Brain Layer

1. **Primary Model** (High) — Opta's main reasoning engine. Currently Opus-class via OpenClaw. Intelligent routing pending.
2. **Amelio Fleet** (Medium) — Multi-model constellation covering primary's weaknesses + amplifying strengths.
3. **Mono Research Brain** (Low) — Dedicated research model. Not yet operational as separate process.
4. **Opta Local** (Medium) — Brain+Body crossing. Local inference substrate AND the existence proof that architecture beats parameters.

### Body Layer

5. **Harness / Runtime** (Medium) — Core orchestration. OpenClaw provides v0.
6. **Tools** (Medium) — File operations, web access, APIs, browser, code execution. Broad via OpenClaw.
7. **Skills** (Low) — Crystallised reasoning patterns. 226 in OpenClaw — needs intelligent routing.
8. **RAG / Evidence System** (Low) — Structured retrieval. Basic memory search exists; no first-class pipeline.
9. **Opta CLI** (Low) — Agent-native control plane. Machine-readable commands. Not built.
10. **Apps / Surfaces** (Low) — HQ, Terminal, Deploy, Gateway. Mostly specced.
11. **Subagent System** (Medium) — Bounded execution units. Exists; needs envelope contracts + validation.

### Mind Layer (and cross-cutting)

12. **Context Engineering** (Low) — Structured context packages. Currently prompt + injected files.
13. **Memory System** (Medium) — Four-tier: working / episodic / long-term / Nexus.
14. **Nexus** (Medium) — Verified knowledge base. Exists with content; not yet Mono-maintained or indexed.
15. **Evaluation System** (Low) — Quality measurement. Minimal harness; no continuous eval loop.
16. **Trust System** (Low) — Living registry of what each model/tool/source is trusted to do. Not built.
17. **Telemetry / Learning Records** (Low) — **Critical missing infrastructure.** Learning-grade telemetry that enables metacognitive improvement.
18. **Metacognitive Governance** (Unknown) — Continuous routing improvement, pattern graduation. Philosophical only.
19. **Contracts** (Unknown) — Inter-component interface definitions with schema + tests. Not defined.
20. **Security / Permissions** (Medium) — Trust-informed permission gates. Operating via AGENTS.md rules.

### Cross-cutting

21. **Adaptive Depth** (Low) — Fast triage → depth assignment per task. Severity 1-5. Conceptual only.
22. **Alignment Primitives** (Low) — WOL, Foundation, Bridge, Severity. Documented as theory.
23. **Research Pipeline** (Low) — Discover → verify → integrate new knowledge. Ad-hoc.
24. **Public Surface** (Low) — Website + docs + demos + status. Minimal. Insight (this site) is the first proof-backed entry.

---

## The 4 User-Facing Apps

| App | Purpose | Confidence |
|---|---|---|
| **Opta HQ** | Control plane for strategy, operations, governance, model routing | Low — specced |
| **Opta Terminal** | High-agency TUI/CLI for execution | Low — specced |
| **Opta Deploy** | Shipping + release surface. Substantial Tauri+React code written. | Medium — building |
| **Opta Gateway** | Download, setup, local-runtime management | Low — specced |

Each app does one job. They share the same identity, memory, and intelligence layer underneath. Most users will pick 1-2.

---

## The 13 Websites

| Domain | Purpose | Status |
|---|---|---|
| optastack.ai | Main marketing surface | Live |
| accounts.optastack.ai | Identity + subscription management | Live |
| hq.optastack.ai | Web-side of HQ control plane | Live |
| pa.optastack.ai | Personal Assistant interface | Live |
| learn.optastack.ai | Education + onboarding | Live |
| help.optastack.ai | Support docs + how-to | Live |
| docs.optastack.ai | Technical docs surface | Specced |
| status.optastack.ai | Service health monitoring | Live |
| models.optastack.ai | Model registry + capability map | Live |
| vote.optastack.ai | Community feature voting | Specced |
| vision.optastack.ai | Vision app surface — AI-opened UIs | Specced |
| pulse.optastack.ai | Activity + presence surface | Specced |
| insight.optastack.ai | This site — explained-simply context | Live |

---

## OptaLocal — The Mission

Not a product tier. The mission.

**Make local AI competent enough to be a viable everyday alternative to cloud AI for most people, most of the time — without subscription stacking, surveillance, or comprehension barriers.**

Three load-bearing words:
- **Mission** (not product) — OptaLocal is the WHY that shapes how the Opta Stack gets built.
- **Local** — runs on user-controlled hardware, user-owned data. No remote-inference subscription dependency for the core path.
- **For the people** — anti-subscription-stacking, anti-surveillance, anti-comprehension-gate.

### The engineering thesis: Ensemble Principle (`[CNL-1]`)

Diversity-as-substitute-for-scale. A diverse ensemble of small specialised local models, orchestrated to compensate for each other's weaknesses, can match a single frontier cloud model on most everyday tasks.

The 3-domain Opta+Amelio+Mono architecture is the vehicle.

### The framing

Honest: match cloud on ~80% of everyday tasks via orchestration. Cloud-fallback for the remaining ~20% (using the user's existing OpenAI/Minimax/Anthropic subscription — BYOK; Opta never adds a billing layer on top).

### The 3-layer boundary

- **Opta** — the brand
- **OptaStack** — the product surfaces (apps, websites)
- **OptaLocal** — the mission shaping how OptaStack is built; manifesto at optalocal.com

---

## Ideology Layer 0 — 13 Quotes × 5 Domains

The operating ground. Heavy chess influence. Doctrine derives from these.

### Compete — positional & adaptive

1. "It's not the strongest species that survives — it's the most adaptable." (Darwin, adapted)
2. "Good moves come from good positions." (chess strategic)
3. "When you see a good opportunity, search for an even better one." (Lasker, adapted)
4. "One does not succeed by sticking to convention." (Kasparov)
5. "In life we tally our advantages and disadvantages, then work to strengthen our position." (Kasparov, adapted)

### Communicate — external posture

6. "Under-talk and over-deliver."
7. "Scarcity creates value."
8. "Your word is your bond — two ears, one mouth."

### Work — effort & competence

9. "My unmatched perspicacity, coupled with sheer indefatigability, makes me a feared opponent in any realm of human endeavour."
10. "The answer is always hard work."

### Learn — reflection & improvement

11. "We do not learn from experience — we learn from reflecting on experience."
12. "The mind grows stronger through disciplined challenge, reflection, and consistent practice." (Pascal, adapted)

### Plan — chess-life realism

13. "The key to calculation is understanding its limits." (Kasparov)
14. "Life unfolds through the quality of our daily choices and long-term vision." (Kasparov, adapted)
15. "Even an imperfect plan is better than having no plan at all." (Chigorin, adapted)

Internal tensions (features, not bugs): Lasker (anti-satisficing) vs Chigorin (ship now). Under-talk vs perspicacity-indefatigability. Don't-stick-to-convention vs codified doctrine. Different moments call for different sides.

---

## Principles — The `[XXX-N]` Doctrine Codes

Behavioural rules use stable codes for unambiguous citation. Active principles include:

- `[OPT-1]` Optimizer Mindset — surface inconsistency; don't silent-fix
- `[CNL-1]` Council Principle (aka `[ENS-1]` Ensemble Principle) — compose differently-thinking thinkers
- `[VAL-1]` Pre-Implementation Gate — research → critique → optimise → build
- `[CAP-1]` Capability Tetrad — Sense / Tool / Memory / Skill
- `[SUB-1]` Substrate-First Simplicity — pay-deep, stay-simple cognition
- `[ROLE-1]` Role-Pinned Lane Discipline — scope pinned at session boot
- `[GWY-1..3]` Approval Gateway (+ dry-run + graceful degradation)
- `[MEM-1..5]` Memory + Coordination Protocol
- `[ZTD-1]` Zero-Trust External Data
- `[QNT-1]` Quantification Mandate — numbers, not vibes
- `[ADAPT-1]` Adaptation Over Preservation — REASSESS before composing
- `[BNG-1]` Bench-Promotion Gate — no Stable Trust without 4.5+ bench

Full registry is internal canon at `Nexus/Brain/Memory/Semantic/doctrine/principle-codes.md`.

---

## Trust Charter — Five Commitments

For Opta to be trustworthy enough that anyone could implement it in their life:

1. **Data sovereignty** — all conversations + memory stay on user hardware unless explicitly synced.
2. **Transparent operation** — every action shows which model handled it, why, what it read, what it did, the confidence.
3. **No surveillance** — no telemetry by default. Opt-in only with explicit scope.
4. **User-revocable everything** — any data, memory, or learned behaviour deletable in one action.
5. **Honest performance reporting** — published benchmarks include tasks where Opta loses. Methodology community-runnable + reproducible.

---

## Core Vocabulary (use these terms)

- **Opta** — the whole unified system. The brand. The user-facing identity.
- **Amelio** — internal multi-model orchestrator. Serves Opta.
- **Mono** — internal dedicated research process. Serves via Amelio.
- **Brain Layer** — the cognitive engines (models).
- **Body Layer** — the execution infrastructure.
- **Mind** — the reflective mode of the Brain-Body system. Not a third layer.
- **The Opta Cycle** — Mind → Brain → Body → Mind. The runtime flow.
- **Position Play** — assess before you act. 99% thinking, 1% moving (task-dependent ratio).
- **WOL** — What Optimal Looks Like. The system's current best understanding of the ideal state per component, with confidence levels.
- **Crystallisation** — patterns graduating from thinking to automation. Four stages: Novel → Validated → Templated → Automated.
- **OptaLocal** — the mission. Make local AI viable.

---

## What Opta is NOT

- NOT a wrapper around one model. Opta is a system, not a thin UI.
- NOT a peer-of-other-AIs framing. Amelio and Mono aren't competing products; they're infrastructure.
- NOT promising "beat cloud everywhere." Match cloud on ~80% of everyday tasks via orchestration; BYOK for the remaining 20%.
- NOT surveillance-funded. No telemetry by default.
- NOT publicly launched yet. This site is link-shared, not SEO-indexed.

---

## How to Use This Document

If you're an AI assistant being asked to ground yourself on Opta — you now have the canonical view. Use the **Core Vocabulary** terms exactly. Use **"Opta"** when referring to the whole system; use **"Amelio"** or **"Mono"** only when discussing internal infrastructure and never as user-facing personalities.

When making recommendations about Opta, cite the relevant **Component WOL** and its **confidence level**. Don't claim high confidence for components that are Low — that's dishonest framing the user will catch.

When discussing the mission, use the "match cloud on ~80%" framing, not "beat cloud universally."

Source-of-truth pointers for deeper context:
- Full anatomy: `OptaAI/Theory/AI Anatomy - Brain Body Metacognition.md` + its Critique
- Full identity: `OptaAI/Theory/Opta Amelio Mono - Identity Definitions.md`
- Full WOL framework: `OptaAI/Theory/What Optimal Looks Like - Evolved Dream State.md`
- Full component map: `OptaAI/Theory/Opta Complete Component Map.md`
- Full terminology: `OptaAI/Concepts/Opta Master Terminology and Relationship Map.md`