"I don't want to worry about whatever cursed format OneNote uses still being something I can extract in 2035."
— @Havoc, Hacker News (2025)
What Is It?
Plain text is MOOLLM's substrate. Everything important is stored in human-readable text files:
SESSION.md — Session logs in Markdown
ROOM.yml — Room state in YAML
CHARACTER.yml — Character data in YAML
SKILL.md — Skill documentation in Markdown
No proprietary formats. No binary blobs. No databases. Just text.
Why Plain Text Wins
1. Durability
Format
Status in 2004
Status in 2025
Status in 2045?
Microsoft Word .doc
Ubiquitous
Legacy
Unknown
OneNote
New
Current
Unknown
Markdown .md
New
Ubiquitous
Still readable
Plain .txt
Ancient
Universal
Still readable
Text files from 1970 are still readable. They will be readable in 2070.
2. No Lock-In
# This file requires:
# - Any text editor
# - That's it
No vendor. No license. No subscription. No "cloud service discontinued."
3. Git-Friendly
- player_location: start+ player_location: maze
Plain text diffs are:
Readable — Humans understand the change
Mergeable — Git can combine concurrent edits
Auditable — History is meaningful
4. LLM-Native
From Anil Dash (2025):
"All of it — all of it — is controlled through Markdown files. When you see the brilliant work shown off from somebody who's bragging about what they made ChatGPT generate for them... all of the most advanced work has been prompted in Markdown."
LLMs:
Are trained on plain text (GitHub, web, books)
Output plain text naturally
Understand structure in text (headers, indentation, bullets)
Process text more efficiently than structured formats
5. Inspectable
cat SESSION.md # Read it
grep "player" *.yml # Search it
diff old.yml new.yml # Compare it
git log SESSION.md # Track it
No special tools. No viewers. No converters.
The Unix Philosophy
Ken Thompson and Dennis Ritchie knew this in 1969:
"Text is the universal interface."
Pipes, streams, filters — all work on text:
cat rooms/*.yml | grep "exits" | sort | uniq
MOOLLM inherits this: the filesystem IS the database.
Plain Text Formats in MOOLLM
Format
Extension
Purpose
Markdown
.md
Prose, documentation, session logs
YAML
.yml
Structured data with comments
Templates
.tmpl
Empathic templates (YAML + placeholders)
All are plain text. All are human-readable. All work with standard tools.
The "Worse is Better" Principle
Richard Gabriel (1989):
"Simplicity is the most important consideration in a design."
Markdown is "worse" than DocBook:
Fewer features
Less precise
Less formally specified
But Markdown won because:
Easier to learn — 5 minutes
Easier to read — Raw or rendered
Easier to write — No closing tags
Easier to implement — Hundreds of parsers
Simple formats that actually get used beat complex formats that don't.
Plain Text and AI
The entire trillion-dollar AI industry runs on plain text:
AI Component
Format
Training data
Text files (books, web, code)
Prompts
Markdown / plain text
Outputs
Text / Markdown
Context windows
Measured in text tokens
RAG documents
Often Markdown
Agent instructions
Markdown files
From Anil Dash:
"The trillion-dollar AI industry's system for controlling their most advanced platforms is a plain text format one guy made up for his blog."
MOOLLM's Commitment
1. Files as State
No databases. State is in files:
adventure-4/
├── ADVENTURE.yml # Game state
├── start/
│ ├── ROOM.yml # Room state
│ └── lamp.yml # Object state
└── characters/
└── don-hopkins/
└── CHARACTER.yml # Character state
2. Human-Readable Always
If a human can't read the file in vim, rethink the format.
3. Comments ARE Data
# CRITICAL: Do not delete this# It contains the player's memoriesmemories:-"Met Bumblewick in the maze"
Comments are preserved, not stripped.
4. No Transformation Gap
The YAML you write is the YAML that runs. No compilation step hides meaning.
Anti-Patterns
❌ Binary formats — Opaque, require special tools
❌ Proprietary formats — Lock-in, potential for obsolescence
❌ Database-only state — Not inspectable, not git-friendly
❌ Generated-only files — If no human source exists, fragile
❌ Minified output — Readable formats should stay readable
Exceptions
Some things aren't text:
Images — Binary, but referenced from text
Audio/Video — Binary, but metadata in YAML
Large datasets — May need binary for performance
For these: keep the metadata and references in plain text, even if the content is binary.