# Video Schema & Script: `localllm` — Building a Local Knowledge Base

* **Target File**: `~/Videos/803.kb.01.md`
* **Target Duration**: ~2 Minutes (120 Seconds)
* **Theme**: Local AI, Emergent Knowledge, Web Search Bridge, and Honest Project Status.
* **Note**: This text is generated by Gemini when help 'za' to do the 'knowledge base' project. (Time-tag: 2026-08-03T19:07:28+08:00)

---

## 🎬 Video Overview & Timeline

| Scene | Time | Focus Area | Key Visuals | Audio / Voiceover Theme |
| :--- | :--- | :--- | :--- | :--- |
| **1** | `00:00 - 00:25` | **Why a Knowledge Base?** | Scattered markdown notes ➔ Clean interconnected wiki | Moving from note graveyards to emergent knowledge. |
| **2** | `00:25 - 00:55` | **The First Step: Web SOS (Search & Crawl)** | Terminal running `py.search` & `py.crawl` | Linking local notes with global web truth. |
| **3** | `00:55 - 01:25` | **Current Project Status (The Honest Truth)** | Terminal code, `DESIGN.md`, CLI commands | Early Phase 0 prototype: rough, CLI-first, but rock-solid policy. |
| **4** | `01:25 - 02:00` | **The Future & Call to Action** | Knowledge Graph render (`knowledge_graph.png`) & MkDocs preview | Local privacy + auto-generated knowledge graphs. |

---

## 📽️ Detailed Scene-by-Scene Script & Visual Schema

### Scene 1: Why a Knowledge Base? (00:00 - 00:25)

* **Visual Description**:
  * *[00:00]* Screen opens on a fast-scroll animation of cluttered folder directories filled with unsorted `.md` files and chat logs.
  * *[00:10]* Transition to a clean, dark-themed diagram showing: `User Notes` + `AI Synthesis` ➔ `Living Wiki`.
* **On-Screen Subtitles & Voiceover**:
  > *"We all have hundreds of scattered Markdown notes, chat logs, and code snippets lying around in folder graveyards. But storing files isn't knowledge.*  
  > *A true Knowledge Base isn't a single 'source of truth'—it's emergent. It takes your raw personal notes, synthesizes them with AI, and turns fragmented thoughts into a living, searchable wiki."*

---

### Scene 2: Why Web Search & Crawl is Step One (00:25 - 00:55)

* **Visual Description**:
  * *[00:25]* Terminal window appears. Type command: `PYTHONPATH=src uv run python -m py.search "happ xray ubuntu networking"`
  * *[00:35]* Animated data flow overlay: DuckDuckGo Search ➔ HTML Web Crawler (`py.crawl`) ➔ Clean Markdown saved in `aispace/search/`.
  * *[00:45]* Diagram showing local `userdocs/` linking directly with web search digests in `aispace/search/`.
* **On-Screen Subtitles & Voiceover**:
  > *"Why is connecting to the global web the very first step of our KB?*  
  > *Because local LLMs can't live in a vacuum. To build accurate wiki pages, the system needs real-world facts.*  
  > *Our search and crawl pipeline—what we call the 'SOS channel'—fetches live web pages, strips out HTML noise, and digests them into clean Markdown, bridging offline local notes with the global internet."*

---

### Scene 3: The Honest Current Status (00:55 - 01:25)

* **Visual Description**:
  * *[00:55]* Show `DESIGN.md` in Neovim/VS Code highlighting `Status: active draft`.
  * *[01:05]* Terminal output running `PYTHONPATH=src uv run python -c "import py.policy"`.
  * *[01:15]* Split screen showing `userdocs/` marked with a 🔒 `READ-ONLY FOR AI` badge, and `aispace/` marked with ✍️ `AI WRITE TARGET`.
* **On-Screen Subtitles & Voiceover**:
  > *"Where does the project stand today? Let's be completely transparent: it's in an early, raw condition.*  
  > *We don't have a flashy GUI yet. It's a Phase 0 prototype built around command-line Python scripts.*  
  > *But the foundational rules are locked down: the AI is strictly forbidden from editing your personal notes in `userdocs/`. It builds and synthesizes strictly inside `aispace/`."*

---

### Scene 4: Knowledge Graphs & What's Next (01:25 - 02:00)

* **Visual Description**:
  * *[01:25]* Show `sample_kg.py` executing in terminal, generating `knowledge_graph.png`.
  * *[01:35]* Display the generated color-coded Knowledge Graph (Entities like `Apollo 11`, `Saturn V`, `Neil Armstrong` with relation edges).
  * *[01:45]* Show `uv run mkdocs serve` running, displaying the multi-source staged static website in browser (`http://127.0.0.1:8000`).
  * *[01:55]* Outro graphic: `localllm — Local First, Privacy First, Emergent Knowledge`.
* **On-Screen Subtitles & Voiceover**:
  > *"Even in this early stage, the pieces are coming together: from text slicing to automated Knowledge Graph extraction and multi-source MkDocs site publishing.*  
  > *100% local, privacy-first, and CLI-driven.*  
  > *Follow along as we build the kernel step by step. Check out `DESIGN.md` in the repo, and let's construct the future of local AI together."*

---

## 🛠️ Production Notes for Creator

1. **Terminal Recording**: Use `vhs` or `asciinema` (or OBS Studio) with a dark theme (e.g. Tokio Night or Catppuccin) and readable 16pt font.
2. **Audio Pacing**: ~170 words total, spoken at a steady, engaging pace (~85-90 words per minute).
3. **Background Music**: Subtle lo-fi or ambient tech synth track at -22dB.

---
*Note: This text is generated by Gemini when help 'za' to do the 'knowledge base' project. Time-tag: 2026-08-03T19:07:28+08:00*


<!-- zatranslate
target = 803.kb.01.cn.md
from = en
to = cn
max_words = 388
-->
