172 lines
4.6 KiB
YAML
172 lines
4.6 KiB
YAML
# ===============================================================================
|
|
# CONTEXT-MAP.yml - Local LLM Agent
|
|
# ===============================================================================
|
|
#
|
|
# Sistema: SIMCO v4.3.0 + NEXUS v4.0
|
|
# Proposito: Mapa central de contexto del proyecto
|
|
# Fecha: 2026-01-24
|
|
#
|
|
# ===============================================================================
|
|
|
|
metadata:
|
|
proyecto: local-llm-agent
|
|
codigo: LLM
|
|
tipo: infrastructure
|
|
nivel_simco: L2-STANDALONE
|
|
version: "0.6.0"
|
|
simco_version: "4.3.0"
|
|
estado: production-ready
|
|
creado: 2026-01-20
|
|
actualizado: 2026-01-24
|
|
|
|
# ===========================================
|
|
# RUTAS DEL PROYECTO
|
|
# ===========================================
|
|
rutas:
|
|
proyecto: projects/local-llm-agent
|
|
gateway: projects/local-llm-agent/apps/gateway
|
|
inference: projects/local-llm-agent/apps/inference-engine
|
|
config: projects/local-llm-agent/config
|
|
monitoring: projects/local-llm-agent/monitoring
|
|
lora: projects/local-llm-agent/lora-adapters
|
|
scripts: projects/local-llm-agent/scripts
|
|
docs: projects/local-llm-agent/docs
|
|
orchestration: projects/local-llm-agent/orchestration
|
|
|
|
# ===========================================
|
|
# ALIASES DEL PROYECTO
|
|
# ===========================================
|
|
aliases:
|
|
"@PROJECT": projects/local-llm-agent
|
|
"@GATEWAY": projects/local-llm-agent/apps/gateway
|
|
"@GATEWAY_SRC": projects/local-llm-agent/apps/gateway/src
|
|
"@INFERENCE": projects/local-llm-agent/apps/inference-engine
|
|
"@INFERENCE_SRC": projects/local-llm-agent/apps/inference-engine/app
|
|
"@CONFIG": projects/local-llm-agent/config
|
|
"@LORA": projects/local-llm-agent/lora-adapters
|
|
"@DOCS": projects/local-llm-agent/docs
|
|
"@ORCH": projects/local-llm-agent/orchestration
|
|
|
|
# ===========================================
|
|
# VARIABLES DEL PROYECTO
|
|
# ===========================================
|
|
variables:
|
|
PROJECT: local-llm-agent
|
|
PROJECT_CODE: LLM
|
|
|
|
# Puertos de servicios
|
|
GATEWAY_PORT: 3160
|
|
INFERENCE_PORT: 3161
|
|
OLLAMA_PORT: 11434
|
|
VLLM_PORT: 8000
|
|
PROMETHEUS_PORT: 9090
|
|
GRAFANA_PORT: 3000
|
|
|
|
# Base de datos (opcional)
|
|
DB_NAME: local_llm_dev
|
|
DB_PORT: 5432
|
|
REDIS_PORT: 6379
|
|
REDIS_DB: 9
|
|
|
|
# ===========================================
|
|
# SERVICIOS
|
|
# ===========================================
|
|
servicios:
|
|
gateway:
|
|
nombre: Gateway API
|
|
puerto: 3160
|
|
framework: NestJS
|
|
lenguaje: TypeScript
|
|
descripcion: API Gateway OpenAI-compatible
|
|
endpoints:
|
|
- "/v1/chat/completions"
|
|
- "/v1/models"
|
|
- "/v1/lora/*"
|
|
- "/mcp/tools/*"
|
|
- "/health"
|
|
|
|
inference_engine:
|
|
nombre: Inference Engine
|
|
puerto: 3161
|
|
framework: FastAPI
|
|
lenguaje: Python
|
|
descripcion: Motor de inferencia con backends multiples
|
|
|
|
backends:
|
|
ollama:
|
|
puerto: 11434
|
|
tipo: cpu
|
|
uso: desarrollo
|
|
vllm:
|
|
puerto: 8000
|
|
tipo: gpu
|
|
uso: produccion
|
|
|
|
monitoring:
|
|
prometheus:
|
|
puerto: 9090
|
|
grafana:
|
|
puerto: 3000
|
|
credenciales: admin/admin
|
|
|
|
# ===========================================
|
|
# MCP TOOLS
|
|
# ===========================================
|
|
mcp_tools:
|
|
- nombre: classify
|
|
descripcion: Clasificar texto en categorias
|
|
tier: small
|
|
- nombre: extract
|
|
descripcion: Extraer datos estructurados
|
|
tier: small
|
|
- nombre: rewrite
|
|
descripcion: Reescribir texto
|
|
tier: main
|
|
- nombre: summarize
|
|
descripcion: Resumir texto
|
|
tier: main
|
|
|
|
# ===========================================
|
|
# TIERS DE PROCESAMIENTO
|
|
# ===========================================
|
|
tiers:
|
|
small:
|
|
max_tokens: 512
|
|
max_context: 4096
|
|
latencia_target_ms: 500
|
|
rate_limit: "60/minute"
|
|
main:
|
|
max_tokens: 2048
|
|
max_context: 16384
|
|
latencia_target_ms: 2000
|
|
rate_limit: "20/minute"
|
|
|
|
# ===========================================
|
|
# HERENCIA DEL WORKSPACE
|
|
# ===========================================
|
|
herencia:
|
|
workspace: workspace-v2
|
|
simco: workspace-v2/orchestration/directivas/simco
|
|
principios: workspace-v2/orchestration/directivas/principios
|
|
templates: workspace-v2/orchestration/templates
|
|
|
|
# ===========================================
|
|
# DOCKER COMPOSE FILES
|
|
# ===========================================
|
|
docker_compose:
|
|
desarrollo: docker-compose.yml
|
|
produccion: docker-compose.prod.yml
|
|
vllm_only: docker-compose.vllm.yml
|
|
monitoring: docker-compose.monitoring.yml
|
|
|
|
# ===========================================
|
|
# CHANGELOG
|
|
# ===========================================
|
|
changelog:
|
|
- version: "1.0.0"
|
|
fecha: 2026-01-24
|
|
cambios:
|
|
- "Creacion inicial del CONTEXT-MAP"
|
|
- "Documentacion de servicios y puertos"
|
|
- "Definicion de tiers y MCP tools"
|