I built slidesmd β€” a tool that watches a folder of .pptx files and auto-generates a single agents.md index that any AI can read.

The problem: I have presentations scattered across folders and no way to search them meaningfully. Slide titles don’t help, file names don’t help, and opening each one to search is tedious. The solution is to extract everything into one structured Markdown file that an AI can query instantly.

How it works

  1. Point slidesmd at a folder containing .pptx files
  2. It extracts metadata β€” title, slide topics, full slide content, to-dos
  3. It generates agents.md β€” a single Markdown index
  4. Ask your AI: β€œWhat was that presentation I did about data pipelines?”

Usage

Set it once, forget it. Auto-updates agents.md whenever you add or remove presentations:

slidesmd watch ~/Documents/Presentations

Manual index

slidesmd index ~/Documents/Presentations

Search from CLI

slidesmd search ~/Documents/Presentations "data pipelines"

Querying with an AI

agents.md is plain Markdown β€” any AI can read it.

Ollama (local, offline)

slidesmd index ~/Documents/Presentations
ollama run llama3 "$(cat ~/Documents/Presentations/agents.md)"

Then ask anything:

>>> What presentations have I done about data pipelines?
>>> Summarise my Iceberg talk in 3 bullet points.

Claude / ChatGPT / Copilot

# macOS β€” copy to clipboard
pbcopy < ~/Documents/Presentations/agents.md

Paste into any chat window and start asking questions.

What agents.md looks like

# Presentations Index

_Auto-generated by slidesmd on 2026-04-04 09:00_

---

## Q3 Sales Review

- **File:** `/Users/you/Documents/Presentations/Q3-Sales-Review.pptx`
- **Slides:** 18
- **Topics:** Introduction, Market Overview, Pipeline, Forecast, Next Steps
- **To-dos:**
  - TODO: Follow up with Nordic accounts by end of month

**Slide content:**

- **Introduction**: What we covered last quarter and what changed
- **Market Overview**: EMEA growth 12% YoY Β· Pipeline up 18%

Image parsing (v0.1.1)

The latest release adds image parsing β€” slides with embedded images are now processed via OCR (pytesseract), with a fallback to Ollama LLaVA for low-confidence results. Ollama is optional and gracefully skipped if not installed. This means diagram-heavy decks are now indexed properly, not just the text slides.

Install

pip install slidesmd

Requires Python 3.10+. Works on macOS, Linux, and Windows. Recurses into subfolders automatically.

Source: github.com/chanukyapekala/slidesmd