# Ask Hormozi

> ask-hormozi

- Skill: `agenticluke/ask-hormozi` (Agent Skill)
- Install (CLI): `npx skillmds@latest add agenticluke/ask-hormozi`
- Raw SKILL.md: https://api.skillmd.com/api/skills/agenticluke/ask-hormozi/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: agenticluke (https://skillmd.com/u/agenticluke)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/agenticluke/ask-hormozi

---


# ask-hormozi

Build a searchable, cited Q&A index from any text corpus. Query with semantic relevance scoring and automatic source citations.

## What It Does

**ask-hormozi** is a local-first citation tool that turns books, podcasts, articles, and transcripts into a searchable knowledge base. You provide text files (or `.vtt` video transcripts), build an index, then query it with natural language. Every answer comes with the exact source, page/chunk, and excerpt.

- No API calls, no vendor lock-in, no rate limits
- TF-IDF semantic search finds the most relevant passages
- Automatic citations: every result shows source file, type, and location
- Plain Python + scikit-learn, runs entirely offline

## Use Cases

- Quote books and podcasts accurately with sources
- Build a FAQ/knowledge index from documentation  
- Create a research tool for your personal learning library
- Cite video transcripts in writing projects
- Search through your own content libraries

## Install

1. **Get the repo** — Clone or download the repository files (`build_index.py`, `query.py`)

2. **Install dependencies**
```bash
pip install scikit-learn numpy
```

3. **Prepare your transcripts** — Create a `transcripts/` folder and add `.txt` (plain text) or `.vtt` (video captions) files
```bash
mkdir transcripts/
# Copy your source files here
```

4. **Build the index**
```bash
python3 build_index.py --transcript-dir ./transcripts
# Creates: index.json
```

## Usage

```bash
python3 query.py "your question here"
```

Results show the top 5 most relevant passages with relevance score and exact source.

### Custom Index Location

```bash
python3 query.py --index-file my_index.json "what was the key insight?"
```

### Rebuild After Adding Transcripts

```bash
python3 build_index.py --transcript-dir ./transcripts --chunk-size 400
```

Flags:
- `--chunk-size N` — Words per chunk (default 500). Lower = more specific, higher = more context.

## How It Works

1. **Index building** reads `.txt` files and `.vtt` transcripts, chunks the text (~500 words per chunk by default), and builds a searchable index with TF-IDF vectors
2. **Search** converts your query to a TF-IDF vector and scores each chunk by cosine similarity
3. **Results** show top matches ranked by relevance (0–1 score), with the source file, chunk number, and excerpt

## Example

```bash
$ python3 query.py "what are the key phases of pricing strategy"

Searching for: "what are the key phases of pricing strategy"
======================================================================

1.
[Text] Chapter 5 - Pricing Strategy (chunk 3)
Score: 0.834

The first phase involves understanding your baseline costs. The second phase 
focuses on market positioning relative to competitors. The third phase tests...

2.
[Video] Podcast Episode 42 - Pricing (Video - chunk 7)
Score: 0.721

Pricing in three steps: understand costs, know your market, test relentlessly...

======================================================================
Found 2 relevant passages. Use these cited excerpts in your research.
```

## Input Formats

- **`.txt`** — Plain text transcripts (UTF-8)
- **`.vtt`** — WebVTT video transcripts (YouTube captions format)

Filenames become source titles automatically.

## Performance

- Indexing: ~10–100 MB of text in <5 seconds
- Queries: ~0.1–1 second per search
- Index size: ~20–30% of original text

## Limits & Tips

- Works best with Latin-script text (English, Spanish, French, etc.)
- TF-IDF may miss synonyms ("expensive" vs "costly") — for advanced semantic search, upgrade to sentence embeddings
- Large indexes (>100k chunks) slow down search — consider splitting into multiple indexes
- Chunk size balances specificity vs. context — start at 500 words, adjust based on your use case

## No Copyright Built In

This tool is **generic and format-agnostic**. You provide your own transcripts, content, and materials. It ships with zero copyrighted content. Use only with:

- Books/audiobooks you own or have rights to analyze
- Your own podcasts, videos, or articles  
- Public-domain or licensed content
- Transcripts you created

## License

MIT. Free to use, modify, share.