Documentation

M2M Protocol Documentation

Overview

M2M (Machine-to-Machine) Protocol is a token-optimized compression scheme for LLM API traffic. Unlike byte-oriented compression (gzip, brotli) that increases token count due to Base64 encoding, M2M achieves 25-40% token reduction through semantic compression.

Documentation Structure

Specification (`spec/`)

Formal protocol specification following IETF RFC conventions:

Document	Description
00-introduction.md	Protocol overview, goals, and scope
01-terminology.md	Definitions and RFC 2119 requirement keywords
02-wire-format.md	Binary/text encoding specification
03-message-types.md	Complete message catalog
04-compression.md	Compression algorithms and mappings
05-session-management.md	Session lifecycle and state machine
06-security.md	Security model and threat mitigations

Guides (`guides/`)

Human-readable documentation for implementers:

Document	Description
quickstart.md	5-minute getting started guide
proxy.md	OpenAI-compatible proxy server

Reference (`reference/`)

API and configuration reference:

Document	Description
error-codes.md	Complete error code reference
configuration.md	Configuration options
abbreviations.md	Key/value abbreviation tables

Examples (`examples/`)

Working examples and test vectors:

Directory	Description
wire-format/	Annotated wire format examples

Quick Reference

Wire Format Prefixes

Algorithm	Prefix	Use Case
Token	`#T1\|`	LLM API payloads (~30% savings)
Brotli	`#BR\|`	Large content (>4KB)
None	(passthrough)	Small content (<100 bytes)

Compression Example

Original (68 bytes):
{"model":"gpt-4o","messages":[{"role":"user","content":"Hello"}]}

Compressed (45 bytes, 34% reduction):
#T1|{"M":"4o","m":[{"r":"u","c":"Hello"}]}

Version History

See CHANGELOG.md for version history.

License

Apache-2.0