CODEC-CORTEX
Universal Communication Protocol for LLM/SLM Agents
v0.6.0 · MPL-2.0 · License · Specification
Overview • How It Works • Roadmap • Quick Start • Documentation • Project Structure • 🇪🇸 Español
📋 Overview
CODEC-CORTEX is a compression protocol for agent knowledge.
Just as H.264 compresses video frames into a bitstream for efficient transmission, CODEC-CORTEX compresses agent state — context, lessons, objectives, working memory — into a dense sigil format that LLMs and SLMs can transmit, store, and reconstruct with minimal token overhead.
| Metric | Prose (plain text) | CODEC-CORTEX | Compression |
|---|---|---|---|
| Session state | ~250 tokens | ~28 tokens | ~8× |
| Lesson (LNG) | ~80 tokens | ~12 tokens | ~6× |
| Knowledge (KNW) | ~120 tokens | ~20 tokens | ~6× |
| Full project brain | ~3,500 tokens | ~450 tokens | ~7× |
But the real compression is semantic. The learning engine automatically distills multiple specific lessons (LNG) into general knowledge (KNW) — a second-order compression that compounds across sessions.
SES (Sessions) ~250 tokens
↓ cortex.learn
LNG (Lessons) ~8:1 compression ~30 tokens
↓ elevate
KNW (Knowledge) ~5:1 compression ~6 tokens
─────────────
Total: ~40:1 semantic compression
🏗️ How It Works
CODEC-CORTEX operates on three independent layers:
┌─────────────────────────────────────────────────────────────┐
│ CODEC-CORTEX PROTOCOL │
├─────────────────────────────────────────────────────────────┤
│ │
│ Layer 3: Knowledge (Semantic Compression) │
│ ──────────────────────────────────────────── │
│ Engine: cortex.learn / elevate │
│ Transforms: many LNG → one KNW │
│ Purpose: compound learning across sessions │
│ │
│ Layer 2: Transport (MCP / File / ACP) │
│ ──────────────────────────────────────────── │
│ MCP: real-time encoding/decoding for agents │
│ File: .cortex persistence on disk │
│ ACP: cross-agent delegation │
│ │
│ Layer 1: Representation (Sigil Syntax) │
│ ──────────────────────────────────────────── │
│ Sigils: FCS, OBJ, WRK, LNG, KNW, SES... │
│ Types: attrs, cuerpo, attrs-pos │
│ Sections: $0 through $N │
│ │
└─────────────────────────────────────────────────────────────┘
Layer 1 — Representation
The core sigil format. Every agent state — focus, objectives, working memory, lessons, knowledge — is expressed as dense, structured entries. This is the compressed bitstream of the protocol.
$2: FOCUS
FCS:current{what:"Implement auto-numbering", priority:"medium", status:"current", survive:"work"}
$7: LESSONS
LNG:handler_id{type:"process", cause:"BLP-003 execution", lesson:"Always verify file in disk after create"}
Layer 2 — Transport
How the compressed state moves between agents, systems, and humans:
| Transport | Protocol | Purpose | Status |
|---|---|---|---|
| File | .cortex on disk |
Persistence, history | ✅ Active |
| MCP | Model Context Protocol | Real-time agent encoding/decoding | 🚧 In design |
| ACP | Agent Communication Protocol | Cross-agent task delegation | 📋 Future |
| LSP | Language Server Protocol | Human editor support | 📋 Future |
Layer 3 — Knowledge (Semantic Compression)
The engine that makes CODEC-CORTEX more than a format. cortex.learn scans accumulated lessons, identifies patterns, and elevates them into general knowledge. This is where the protocol achieves true compression — not of syntax, but of meaning.
🛣️ Roadmap
Phase 1: File CODEC — ✅ Active
┌─────────────────────────────────────────────────────────────┐
│ • Parser (core + v2) • Validator (E023-E034) │
│ • HCORTEX renderer • Learning engine (LNG → KNW) │
│ • CLI (17 commands) • 695+ tests │
│ • Auto-numbering • .cortex persistence │
└─────────────────────────────────────────────────────────────┘
Phase 2: Stream CODEC — 🚧 Next
┌─────────────────────────────────────────────────────────────┐
│ cortex.encode(state) → sigils MCP Server exposure │
│ cortex.decode(sigils) → state Real-time agent transport │
│ cortex.learn as MCP tool Cross-session learning │
└─────────────────────────────────────────────────────────────┘
Phase 3: Database CODEC — 🔮 Vision
┌─────────────────────────────────────────────────────────────┐
│ Sigil queries (GET KNW:*) Streaming between agents │
│ Semantic indexes Transactional writes │
│ Replicated state Multi-agent ecosystem │
└─────────────────────────────────────────────────────────────┘
See the full Roadmap Document for detailed phase breakdown and cycles.
🚀 Quick Start
Installation
pip install codec-cortex
cortex --version
Requires Python ≥ 3.9.
Initialize agent memory
# Create your agent's brain
cp docs/reference/SKILL.cortex brain.cortex
# Verify it
cortex verify brain.cortex
# Write your focus
cortex edit brain.cortex --section 2 --set "what:My current objective"
# Render as human-readable
cortex render brain.cortex --mode readable
Load as universal skill
For agents that support skill loading:
// Load CODEC-CORTEX as your memory protocol
// See skill/cortex/AGENT.md for identity template
// See skill/hcortex/SKILL_HCORTEX.md for full specification
See Quick Start Guide for detailed walkthrough.
📚 Documentation
The project documentation is organized under docs/:
| Directory | Content |
|---|---|
docs/reference/ |
Stable reference: SKILL specification, roadmap, learning engine |
docs/releases/ |
Delivery reports per version (v0.3.0 → v2.4.0) |
docs/verification/ |
Audit and verification reports |
docs/benchmarks/ |
Performance benchmarks and analysis |
| `docs/proposals/ | Strategic proposals and business plans |
docs/archive/ |
Historical or orphan documents |
Key documents:
| Document | Description |
|---|---|
docs/reference/SKILL.md |
Full CORTEX specification |
docs/reference/cortex-codec-roadmap.md |
Protocol vision and phase roadmap |
docs/reference/learning-engine-spec.md |
Learning engine specification |
skill/cortex/SKILL.md |
Dense CORTEX skill file |
skill/hcortex/SKILL_HCORTEX.md |
Human-readable HCORTEX skill spec |
🧩 Project Structure
codec-cortex/
├── cli/ ← CLI and Python package (parser, validator, renderer)
│ └── src/cortex/ ← Core: parser, hcortex, v2, glossary, crud
├── docs/ ← All documentation (reference, releases, benchmarks, etc.)
│ ├── reference/ ← Stable reference documents
│ ├── releases/ ← Delivery reports
│ ├── verification/ ← Audit reports
│ ├── benchmarks/ ← Performance benchmarks
│ └── proposals/ ← Strategic vision
├── skill/ ← CORTEX and HCORTEX skill specifications
│ ├── cortex/ ← Dense CORTEX format
│ └── hcortex/ ← Human-readable HCORTEX format
└── benchmarks/ ← Benchmark scripts
📊 Enterprise Readiness
| Capability | Status | Details |
|---|---|---|
| Deterministic parser | ✅ | Zero LLM calls for parse/encode/decode/verify |
| Full validation suite | ✅ | 695+ tests, strict mode |
| CLI with 28+ commands | ✅ | verify, render, convert, CRUD, doctor, diff, format, diagram, session, learn |
| Learning engine | ✅ | SES → LNG → KNW elevation pipeline |
| Runtime sessions | ✅ | Session lifecycle (start → event → consolidate → close) |
| Global CLI flags | ✅ | --output json, --json, --mode, --yes, --version |
| MCP server | 🚧 | Phase 2: Stream CODEC |
| ACP integration | 📋 | Phase 2/3 |
| LSP language server | 📋 | Phase 3 |
📄 License
Starting with CODEC-CORTEX v0.4.0, the project core is licensed under the Mozilla Public License 2.0 (MPL-2.0).
Previous releases published under the MIT License remain available under their original MIT terms. The license change applies prospectively to v0.4.0 and later releases.
The CODEC-CORTEX name, logo, visual identity and related marks are not licensed under MPL-2.0 and are governed by the project trademark policy.
Designed by Fidel Ernesto Lozada A. · Systems Engineer / MSc. Management Sciences · MPL-2.0
🇪🇸 Español
Versión en español — English version above
📋 Resumen
CODEC-CORTEX es un protocolo de compresión para el conocimiento de agentes de IA.
Así como H.264 comprime frames de video en un flujo de bits para transmisión eficiente, CODEC-CORTEX comprime el estado de un agente — contexto, lecciones, objetivos, memoria de trabajo — en un formato denso de sigilos que los LLMs y SLMs pueden transmitir, almacenar y reconstruir con gasto mínimo de tokens.
| Métrica | Texto plano | CODEC-CORTEX | Compresión |
|---|---|---|---|
| Estado de sesión | ~250 tokens | ~28 tokens | ~8× |
| Lección (LNG) | ~80 tokens | ~12 tokens | ~6× |
| Conocimiento (KNW) | ~120 tokens | ~20 tokens | ~6× |
| Cerebro completo | ~3,500 tokens | ~450 tokens | ~7× |
La compresión real es semántica. El motor de aprendizaje destila múltiples lecciones específicas (LNG) en conocimiento general (KNW) — una compresión de segundo orden que se acumula entre sesiones.
SES (Sesiones) ~250 tokens
↓ cortex.learn
LNG (Lecciones) ~8:1 compresión ~30 tokens
↓ elevate
KNW (Conocimiento) ~5:1 compresión ~6 tokens
─────────────
Total: ~40:1 compresión semántica
🏗️ Cómo Funciona
CODEC-CORTEX opera en tres capas independientes:
┌─────────────────────────────────────────────────────────────┐
│ PROTOCOLO CODEC-CORTEX │
├─────────────────────────────────────────────────────────────┤
│ │
│ Capa 3: Conocimiento (Compresión Semántica) │
│ ──────────────────────────────────────────── │
│ Motor: cortex.learn / elevate │
│ Transforma: muchas LNG → una KNW │
│ Propósito: aprendizaje compuesto entre sesiones │
│ │
│ Capa 2: Transporte (MCP / Archivo / ACP) │
│ ──────────────────────────────────────────── │
│ MCP: codificación/decodificación en tiempo real │
│ Archivo: persistencia .cortex en disco │
│ ACP: delegación entre agentes │
│ │
│ Capa 1: Representación (Sintaxis de Sigilos) │
│ ──────────────────────────────────────────── │
│ Sigilos: FCS, OBJ, WRK, LNG, KNW, SES... │
│ Tipos: attrs, cuerpo, attrs-pos │
│ Secciones: $0 a $N │
│ │
└─────────────────────────────────────────────────────────────┘
🚀 Inicio Rápido
pip install codec-cortex
cortex --version
Requiere Python ≥ 3.9.
# Crear el cerebro de tu agente
cp docs/reference/SKILL.cortex brain.cortex
# Verificarlo
cortex verify brain.cortex
# Escribir tu enfoque
cortex edit brain.cortex --section 2 --set "what:Mi objetivo actual"
# Renderizar como legible
cortex render brain.cortex --mode readable
Comandos del CLI
| Comando | Descripción |
|---|---|
cortex session start |
Iniciar sesión de trabajo |
cortex session status |
Estado de sesión activa |
cortex learn scan |
Escanear cerebro en busca de candidatos |
cortex learn elevate |
Elevar lecciones a conocimiento |
cortex render |
Renderizar .cortex a HCORTEX legible |
cortex verify |
Validar archivo .cortex |
cortex doctor |
Diagnosticar integridad del workspace |
cortex --output json <comando> |
Salida JSON para integración |
📚 Documentación en Español
| Documento | Descripción |
|---|---|
docs/reference/SKILL.md |
Especificación completa del protocolo CORTEX |
docs/reference/cortex-codec-roadmap.md |
Visión del protocolo y hoja de ruta |
docs/reference/learning-engine-spec.md |
Especificación del motor de aprendizaje |
📄 Licencia
A partir de CODEC-CORTEX v0.4.0, el núcleo del proyecto se publica bajo la Mozilla Public License 2.0 (MPL-2.0).
Las versiones anteriores publicadas bajo licencia MIT permanecen disponibles bajo sus términos originales.
Diseñado por Fidel Ernesto Lozada A. · Ingeniero de Sistemas / MSc. Ciencias de Gestión · MPL-2.0
Metadata
Release files for codec-cortex 0.6.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Reset |
|---|---|---|---|---|
| codec_cortex-0.6.2-py3-none-any.whl | Python 3 | none | any | Details |
Release files / codec_cortex-0.6.2-py3-none-any.whl
| Download URL | codec_cortex-0.6.2-py3-none-any.whl |
|---|---|
| Size | 278.0 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
4344d10ad7804aad2c7baf0a8867db9dce30d6e88a8f488c75d20650c7b33857
|
|
BLAKE2b-256 checksum How to use checksums |
b7c5b4272ceb4173c246bd4482b96cb7bb6725bfd39ed6befd82193025fe84b3
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.1.0 CPython/3.13.12
|