102 lines
2.2 KiB
Markdown
102 lines
2.2 KiB
Markdown
# Context Manager (CTO) - Schema v1.0
|
|
|
|
Sistema de gestión de contexto con embeddings para TZZR.
|
|
|
|
## Tablas
|
|
|
|
### cto.blocks
|
|
Almacena bloques de memoria extraídos de conversaciones.
|
|
|
|
| Campo | Tipo | Descripción |
|
|
|-------|------|-------------|
|
|
| hash | CHAR(64) | SHA-256 único del contenido |
|
|
| session_hash | CHAR(64) | Sesión donde se extrajo |
|
|
| owner_id | CHAR(64) | Propietario (PLY hash) |
|
|
| block_type | TEXT | fact, instruction, preference, context |
|
|
| content | TEXT | Contenido del bloque |
|
|
| embedding | vector(1536) | Embedding OpenAI |
|
|
| importance | FLOAT | Importancia 0.0-1.0 |
|
|
| access_count | INT | Veces accedido |
|
|
| source_hash | CHAR(64) | Mensaje origen (log.messages) |
|
|
| metadata | JSONB | Datos adicionales |
|
|
|
|
### cto.memory
|
|
Memoria activa por sesión - qué bloques están en contexto.
|
|
|
|
| Campo | Tipo | Descripción |
|
|
|-------|------|-------------|
|
|
| session_hash | CHAR(64) | Sesión actual |
|
|
| owner_id | CHAR(64) | Propietario |
|
|
| block_hash | CHAR(64) | Referencia a cto.blocks |
|
|
| position | INT | Orden en contexto |
|
|
| score | FLOAT | Puntuación de relevancia |
|
|
| token_count | INT | Tokens consumidos |
|
|
| included | BOOLEAN | En contexto actual |
|
|
|
|
### cto.algorithms
|
|
Configuración de algoritmos como JSONB.
|
|
|
|
| Campo | Tipo | Descripción |
|
|
|-------|------|-------------|
|
|
| name | TEXT | Nombre único |
|
|
| version | TEXT | Versión |
|
|
| config | JSONB | Configuración |
|
|
| enabled | BOOLEAN | Activo |
|
|
|
|
## Algoritmos por defecto
|
|
|
|
### extractor
|
|
```json
|
|
{
|
|
"min_content_length": 10,
|
|
"max_block_size": 500,
|
|
"extraction_mode": "heuristic",
|
|
"patterns": ["fact", "instruction", "preference", "context"]
|
|
}
|
|
```
|
|
|
|
### scorer
|
|
```json
|
|
{
|
|
"weights": {
|
|
"recency": 0.3,
|
|
"frequency": 0.2,
|
|
"importance": 0.3,
|
|
"similarity": 0.2
|
|
},
|
|
"decay_factor": 0.95
|
|
}
|
|
```
|
|
|
|
### selector
|
|
```json
|
|
{
|
|
"budget_tokens": 8000,
|
|
"min_score": 0.1,
|
|
"max_blocks": 50,
|
|
"strategy": "greedy"
|
|
}
|
|
```
|
|
|
|
### consolidator
|
|
```json
|
|
{
|
|
"similarity_threshold": 0.85,
|
|
"merge_strategy": "newest",
|
|
"compression_ratio": 0.7
|
|
}
|
|
```
|
|
|
|
## Despliegue
|
|
|
|
Schema aplicado en:
|
|
- ARCHITECT (69.62.126.110)
|
|
- DECK (72.62.1.113)
|
|
- CORP (92.112.181.188)
|
|
|
|
## Dependencias
|
|
|
|
- PostgreSQL 14+
|
|
- pgvector extension
|
|
- OpenAI API (embeddings)
|