Why MOL?¶

The Problem¶

Every AI pipeline today looks like this:

# Python: 50+ lines of glue code
from langchain import ...
from chromadb import ...
import logging

logging.basicConfig(level=logging.DEBUG)
logger = logging.getLogger(__name__)

loader = TextLoader("data.txt")
docs = loader.load()
logger.debug(f"Loaded {len(docs)} docs")

splitter = RecursiveCharacterTextSplitter(chunk_size=512)
chunks = splitter.split_documents(docs)
logger.debug(f"Split into {len(chunks)} chunks")

embeddings = OpenAIEmbeddings()
vectors = embeddings.embed_documents([c.page_content for c in chunks])
logger.debug(f"Embedded {len(vectors)} vectors")

db = Chroma.from_documents(chunks, embeddings)
results = db.similarity_search("query", k=5)
logger.debug(f"Found {len(results)} results")

You're writing glue code, not logic. And you added logger.debug() everywhere just to see what happened.

The MOL Solution¶

"data.txt"
  |> load_text
  |> chunk(512)
  |> embed
  |> store("index")
  |> retrieve("query", 5)

That's it. Five lines. Auto-tracing shows you every step automatically:

┌─ Pipeline Trace ──────────────────────────────────────
│ 0.  input               ─  Text("data.txt")
│ 1.  load_text        0.1ms  → Document(...)
│ 2.  chunk(..)        0.0ms  → List<3 chunks>
│ 3.  embed            0.2ms  → List<3 embeddings>
│ 4.  store(..)        0.1ms  → VectorStore("index")
│ 5.  retrieve(..)     0.1ms  → List<5 docs>
└─ 5 steps · 0.5ms total ───────────────────────────

No logging setup. No debug prints. It's built into the language.

How MOL Compares¶

Feature	Python	Elixir	F#	Rust	MOL
Pipe operator `\\|>`	No	Yes	Yes	No	Yes
Auto-tracing	No	No	No	No	Yes
AI domain types	No	No	No	No	Yes
RAG built-in	No	No	No	No	Yes
Reads like English	No	Partial	No	No	Yes
Transpiles to JS+Python	N/A	No	No	No	Yes

Key Advantages¶

1. Pipeline Visibility¶

Other languages with |> (Elixir, F#) still make you add logging manually. MOL auto-traces any pipe chain with 3+ stages.

2. English-Like Syntax¶

let name be "MOL"
set version to 1.0
if name is "MOL" then
  show "Welcome!"
end

No curly braces. No semicolons. It reads like pseudocode.

3. AI-Native Types¶

let thought be Thought("idea", confidence: 0.9)
let doc be Document("content", source: "wiki")
let vec be Embedding("hello", model: "default")

These aren't dicts or classes — they're language-level types with built-in behavior.

4. Safety Built In¶

guard score >= 0 : "Score cannot be negative"
guard len(items) > 0 : "List must not be empty"

Guard assertions halt with clear messages. No unchecked nulls or silent failures.

5. Write Once, Run Anywhere¶

mol run app.mol              # Run directly
mol transpile app.mol -t py  # Generate Python
mol transpile app.mol -t js  # Generate JavaScript
mol build app.mol            # Compile to browser HTML

One .mol file targets three runtimes.

Who Is MOL For?¶

AI engineers building RAG pipelines who want visibility
Data scientists who want readable data transformations
Educators teaching programming with natural-syntax language
Startups building AI products that need to move fast
Anyone tired of writing logging boilerplate

Try It Now¶

pipx install mol-lang
mol repl

Or visit the online playground — no installation required.