Get Started

Learn the Notation

Agent Notation is a human-readable shorthand for LLM prompts. It replaces verbose English with compact symbols and abbreviations, cutting token costs by up to 63% while preserving full meaning.

Operators

Common Abbreviations

7 Compression Rules

1. Drop filler — remove articles, copulas, hedging words
2. Abbreviate — use standard abbreviations (function → fn)
3. Use operators — replace verbal relationships with symbols (returns → →)
4. Annotate types — use colon syntax (x:int)
5. Comma-separate — list parallel operations
6. Key-value pairs — for configuration (tone:formal)
7. Conditionals — use ? and | for branches

Worked Example

English:

Write a JavaScript async function that fetches user data from the database,
validates the response, and returns a formatted JSON document, or returns
an error message if validation fails.

Step by step:

1. Language prefix: js
2. Abbreviate: async fn, usr, db, val, res, fmt, doc, err, msg
3. Add operators: → for flow, | for alternative

Result:

js async fn: usr db→val(res)→fmt doc:json | err msg

Try it in the Playground →

CLI

Install

bashnpm install -g @agentnotation/cli

Or use it via npx without installing:

bashnpx @agentnotation/cli compress "Write a Python function that filters even numbers"

Commands

compress — English to notation

bash# From argument
anot compress "Validate user input and return formatted errors"

# From stdin
echo "Validate user input and return formatted errors" | anot compress

# JSON output
anot compress --json "Validate user input and return formatted errors"

# Batch mode (one prompt per line)
cat prompts.txt | anot compress --batch

expand — Notation to English

bashanot expand "val(usr input)→fmt err"

# JSON output
anot expand --json "val(usr input)→fmt err"

# Batch mode
cat notation.txt | anot expand --batch

validate — Check notation syntax

bashanot validate "fn(x:int)→int"
# ✓ Valid Agent Notation

anot validate "fn((x)"
# error: Unbalanced parentheses

lint — Best-practice suggestions

bashanot lint "function(x)→return y"
# warning: "function" can be abbreviated to "fn"
# warning: "return" can be abbreviated to "ret"

stats — Compare token costs

bashanot stats \
  --english "Write a Python function that filters even numbers" \
  --notation "py fn: filter even nums"
# English: 9 tokens | Agent Notation: 7 tokens | Savings: 22%

VS Code Extension

Install

bashcd vscode-extension
npm install && npm run package
# Install the .vsix in VS Code: Extensions → ⋯ → Install from VSIX

Features

• Syntax highlighting — operators, abbreviations, and comments in .tforge and .tf files
• Inline diagnostics — validation errors and lint warnings appear as you type
• Hover tooltips — hover any abbreviation to see its expansion
• Autocompletion — suggests abbreviations and operators as you type
• Token count — live token count in the status bar
• Compress / Expand commands — right-click or use the command palette

AI Tool Integrations

Teach any AI tool to read and write Agent Notation. Pick your tool below for copy-pasteable setup instructions.

You understand Agent Notation, a compact shorthand for LLM prompts.

Key operators: → (flow/returns), > (sequence), ? (conditional), | (else),
&& (AND), || (OR), # (state prefix), ~ (duration), @ (target), × (group-by),
+ (combine), ! (negate), : (type), [] (list).

Common abbreviations: fn=function, ret=return, val=validate, ctx=context,
cfg=configuration, req=request, res=response, msg=message, err=error,
auth=authentication, db=database, usr=user, doc=document, fmt=format,
py=Python, js=JavaScript, ts=TypeScript.

When you receive notation, expand it to understand the full intent.
When asked to compress, use these rules to produce compact notation.

Full reference: https://agentnotation.dev/llms-full.txt