# Pepa

<table>
    <tr>
        <td>AI Assisted</td>
    </tr>
</table>

> > ***Pepa** — a self-hosted, octopus-inspired, cognitive infrastructure **blueprint** to help the elderly preserve memory, organize knowledge, and maintain agency through **deterministic** AI-assisted systems.*
>

---

### 1. Purpose

Pepa is a personal AI cognitive assistant designed to help people remember things, organize knowledge, and maintain independence as they age.

The project explores how a **self-hosted AI system can function as long-term cognitive support infrastructure**, combining automation, memory systems, and local artificial intelligence.

Pepa begins as a personal system but is designed so others — particularly older adults and their caregivers — can eventually benefit from the same architecture.

---

### 2. Vision

Computers have long augmented human physical abilities.

Pepa explores how **AI can supplement and support decaying human cognitive abilities** — particularly memory, organization, and decision support.

The long-term vision is a **personal cognitive infrastructure** that:

- captures life information
- organizes it intelligently
- recalls it when needed
- assists with daily tasks
- evolves with the user over time

---

### 3. Design Philosophy

1. **Deterministic Core** — Orchestration remains predictable; AI assists but does not control.
2. **Human-Centered AI** — Pepa supplements cognition, not replace it.
3. **Data Ownership** — All personal data remains under the user’s control.
4. **Longevity & Maintainability** — Designed to survive hardware and software failures, evolution and lifecycle.
5. **Calm Technology** — Everything must work together without causing a "system overload" for the elderly.
6. **Open Knowledge** — Architecture and lessons are documented openly for reproducibility and improvement.

---

### 4. Octopus Architecture

Pepa mimics nature, specifically the **Octopus Model** where:

- the **Head** (central core) performs deterministic orchestration coordinating subsystems, managing memory, and routing requests;
- the **Arms** (*we don't call them tentacles*) are specialized subsystems that perform specific functions independently but are supervised by the head;
- the **Beak** is the hardware/software infrastructure that runs (feeds) the entire system: sensors, interfaces, compute, storage, networking, UPS, monitoring; it connects the parts like a nervous system.

| Arm          | Function                                                                                           |
|--------------|----------------------------------------------------------------------------------------------------|
| Automation   | Home Assistant, Node-RED, n8n; “sensory arm” that interacts with the real world                    |
| Memory       | ChromaDB, PostgreSQL, MemoriesDB, vector + graph memory stores                                     |
| Knowledge    | Document storage, tech article queues, structured information                                      |
| AI Reasoning | Ollama, local LLMs, embeddings, vector + graph search                                              |
| Coding Agent | Could be a bliss or a loose loaded gun; will probably require a subscription and very close supervision |
| Others       | To be determined as needed                                                                         |

**Benefits of the Octopus Model:**

- **Resilience** — Subsystems can fail without collapsing the whole system
- **Modularity** — Each piece evolves independently
- **Maintainability** — Easier to debug and understand
- **Scalability** — New capabilities can be added as new arms, head or beak improvements

**Core Principle:** AI assists reasoning; deterministic orchestration maintains reliability and predictability. _The exact opposite of the latest AI lobster claw craze. And octopii eat lobsters._

---

### 5. Guiding Constraints

Pepa is designed within **real-world limitations**:

- modest hardware, software and subscription budgets
- homelab environments in older houses
- unpredictable and expensive energy supply
- aging hardware and upstream software dependencies
- incremental improvement and experimentation

These constraints ensure Pepa is **practical, reproducible, and maintainable**.

---

### 6. What Pepa Is Not

- **Not an AGI project** — Focuses on practical cognitive assistance, not Artificial General Intelligence.
- **Not a chatbot** — Conversation is only an interface; memory and task management are primary.
- **Not a robot or mechanical assistant** — Pepa is software and systems-based, not a physical companion or humanoid device.
- **Not cloud-dependent** — Designed to function primarily with self-hosted infrastructure.
- **Not a monolithic AI agent** — Uses octopus-inspired modular subsystems; deterministic orchestration remains centralized.
- **Not a commercial product** — Open exploration of personal cognitive infrastructure, not a startup.
- **Not perfect** — Being built in a modest homelab by its own aging user so incremental improvements and failures are expected and documented.

---

### 7. Invitation to Collaborate

Pepa is an open exploration into the future of personal AI assistance.

We welcome developers, engineers, and thinkers interested in:

- *Supporting the elderly they love, or will one day become themselves*
- AI agents
- memory systems
- personal knowledge infrastructure
- self-hosted AI
- assistive technologies

Collaborators are invited to contribute ideas, critique, and improvements, with the understanding that Pepa is **experimental, modular, and human-centered**.
[Contact the editor](mailto:editor@ai4aging.org).

### 8. Origin Story / Problem Statement

Pepa began as a personal solution to a universal problem: **human memory and cognitive capacity decline with age**. At 65, I recognized firsthand that even the most organized mind can forget important details, miss patterns, or struggle with complex daily tasks. I wanted a system that would **support memory, maintain independence, and adapt over time**, rather than relying on temporary digital notes or commercial assistants.

The arrival of **local language models and open source AI tools** made the project technically and financially achievable at my pay grade. Pepa could now combine **automation, structured memory storage, and reasoning engines** to create a self-hosted cognitive ecosystem — an AI system that **supplements human memory and decision-making without replacing the human**.

In short, Pepa is not just software and cabling. It is the **practical embodiment of a personal cognitive assistant**, born from necessity, informed by decades of engineering and automation experience, and designed to **support the elderly while respecting autonomy, privacy, and human-centered principles**.