Runtime Current State

Estado actual real

Control plane

Hostname: ubuntu-ialab
IP principal: 192.168.1.30
Rol: primary-control-plane
Hermes Enterprise status endpoint: 192.168.1.30:8095 (GET /hermes/status)
Servicios: ailab-gateway (:8008), ailab-router (:8083), ailab-live-api (:8084), ailab-hermes-status (:8095)

Observability

Prometheus: 192.168.1.40:9090 — source of truth
Grafana: 192.168.1.40:3000 — visualization layer (15 dashboards, TIER 1 + TIER 2)
Alert rules: 19 reglas activas, todas health=ok
Métricas: 100+ ailab_* covering profiles, latency, tools, memory, quality, streaming, GPU, SLO, governance, sensor fusion
Loki: log layer
Los sensores del runtime se alimentan de Prometheus y del API de LM Studio.

Inference backend activo

GPU: RX9070
Host: 192.168.1.50
Estado: online
Fuente operativa: Prometheus GPU exporter + LM Studio API

Inference backend inventariado

GPU: RX7900XT
Host: 192.168.1.60
Estado: expected_offline / inventory
No activo
No routable
No debe generar métricas inventadas

Storage

Runtime vivo: /opt/ai-lab
Runtime data: /opt/ai-lab-data
Modelos: /mnt/ai-models
Archives históricos: /mnt/opencode/ai-lab-archives

Enterprise layer (Hermes Enterprise Core)

Hermes Enterprise Core is deployed as a read-only declarative layer exposing AI-LAB’s enterprise semantics via GET /hermes/status on port :8095. All components are in validation-only mode — no active enforcement, no hooks executing.

Component	Status	Description
SOUL	✅ Implemented	Ontological seed: truth_model, boundaries, protocols, domains, identity, schema
Capability Registry	✅ Implemented	Runtime capability catalog with cross-validation (24 capabilities)
Operator Registry	✅ Implemented	12 deep validations per operator (IDs, capabilities, MCP, protocols, execution_mode, domains, forbidden_actions)
Hook Registry	✅ Implemented	Lifecycle hooks skeleton (all `enabled: false, mode: declarative_only`)
MCP Registry	✅ Implemented	Declared MCP servers (prometheus, marketplace-mcp as planned, no active tools)
Dynamic Governance	✅ Implemented	4 modes (NORMAL/ELEVATED/DEGRADED/LOCKDOWN), 6 trigger signals, anti-flapping 30s, capability-governance matrix
Status Endpoint	✅ Implemented	GET /hermes/status → :8095, 14 response blocks, always-on 200

185 tests PASS. Checkpoint: CP-HERMES-ENTERPRISE-CORE-01.

Knowledge Base (AnythingLLM Enterprise)

AnythingLLM Enterprise provides the RAG knowledge layer for AI-LAB. All workspaces are configured with LM Studio backend (192.168.1.50:1234) and text-embedding-multilingual-e5-small embedder.

Resource	Value
Host	192.168.1.50:3001
Active workspaces	7
Total vectors	1304
Embedder	text-embedding-multilingual-e5-small (Q8_0)
Document corpus	Architecture (84 canonical), Evidence reports (53), Marketplace (7), Observability+IDS (2), Runbooks+Stack (8), MCP+A2A (19)
RAG validation	✅ E2E 100%
Baseline	Frozen at CP-ANYTHINGLLM-ENTERPRISE-04-COMPLETE

Qué cambió con 30I

Antes:

LLM + routing + prompts

Después:

Runtime observacional cognitivo
→ Prometheus-backed evidence
→ Sensor fusion
→ Evidence-bound reporting
→ GPU operational summaries
→ source_of_truth / freshness / confidence

Checkpoints relevantes

Bloques 28–30 — Agentes, streaming, SLO enforcement, sensor fusion

CP-28.1-PLANNER-RUNTIME-SKELETON-STABLE
CP-28.2-READONLY-EXECUTOR-STABLE
CP-28.2-B-READONLY-BURNIN-STABLE
CP-28.3-SANDBOX-WRITE-STABLE
CP-28.3-B-SANDBOX-WRITE-BURNIN-STABLE
CP-29.2-B-STREAMING-BURNIN-STABLE
CP-29.3-THREE-MODEL-RUNTIME-STABLE
CP-29.4-SLO-ENFORCEMENT-STABLE
CP-29.4.4-ERROR-TAXONOMY-STABLE
CP-29.4.4-C-SLO-HEALTH-ENDPOINT-STABLE
CP-29.4.4-D-PARALLEL-TOOLCALL-HARDENING-STABLE
CP-30A-RUNTIME-STATE-FOUNDATION-STABLE
CP-30B-MODEL-STATE-AWARE-STABLE
CP-30E-GOVERNANCE-VISIBILITY-STABLE
CP-30F-ROUTE-SEMANTICS-STABLE
CP-30G-OPERATIONAL-REPORTING-STABLE
CP-30H-RUNTIME-EVIDENCE-ENFORCEMENT-STABLE
CP-30H.1-UNIVERSAL-EVIDENCE-GUARD-STABLE
CP-30H.2-RUNTIME-CONTEXT-INJECTION-STABLE
CP-30I-RUNTIME-SENSOR-FUSION-STABLE
CP-30I-B-SENSOR-FUSION-HARDENED-STABLE
CP-30I-C-SENSOR-SUMMARY-EXPOSURE-STABLE
CP-30I-D-SENSOR-SEMANTICS-NORMALIZED-STABLE
CP-30I-E-OPERATIONAL-RESPONSE-FORMATTING-STABLE
CP-30I-F-RUNTIME-COGNITIVE-COMPRESSION-STABLE
CP-30I-F0-RUNTIME-MODEL-ROUTING-CLEANUP-STABLE
CP-30I-G-RUNTIME-GROUNDING-STABLE

Bloques 31–36 — Madurez semántica, cognición, precisión

CP-OBS-31A-OBSERVABILITY-SOURCE-OF-TRUTH-STABLE
CP-OBS-31A.1-PROMETHEUS-AUTHORITY-AUDIT-STABLE
CP-OBS-31A.2-GRAFANA-DRIFT-AUDIT-STABLE
CP-OBS-31A.3-RUNTIME-OBSERVABILITY-ALIGNMENT-STABLE
CP-OBS-31A.4-OBSERVABILITY-REMEDIATION-PLAN-STABLE
CP-OBS-31A.5-EXECUTOR-STABLE
CP-31B-RUNTIME-SEMANTIC-MATURITY-STABLE
CP-31B-HF1-OPENCODE-CONTEXT-ALIGNMENT-STABLE
CP-31C-OPERATIONAL-REPORTING-DISCIPLINE-STABLE
CP-35C-LIVE-AUTHORITY-BACKED-COGNITION-STABLE
CP-35D-OPERATIONAL-FAST-PATH-STABLE
CP-35D-HF1-FASTPATH-ROUTING-PRIORITY-STABLE
CP-36A-OPERATIONAL-INCIDENT-INTELLIGENCE-STABLE
CP-DEV-36X-CODEBASE-MEMORY-INTEGRATION-STABLE
CP-DOC-36X-GITNEXUS-STRUCTURAL-COGNITION-STABLE
CP-DOC-36X-SPANISH-LOCALIZATION-STABLE
CP-36B-RUNTIME-PRECISION-MODE-STABLE

Bloques 37–40 — Estabilidad, release hardening, SLO drift watch

CP-38A-RUNTIME-DEEP-AUDIT-01-STABLE
CP-38B-GATEWAY-SHUTDOWN-GRACEFUL-01-STABLE
CP-38C-GITNEXUS-NAPI-ERROR-TRIAGE-01-STABLE
CP-38D-RUNTIME-STABILITY-SNAPSHOT-01-STABLE
CP-39A-OPENCODE-GATEWAY-CONTRACT-HARDENING-01-STABLE
CP-39B-RUNTIME-OBSERVABILITY-ALERTS-01-STABLE
CP-39C-COGNITIVE-HEALTH-FOLLOWUP-01-STABLE
CP-39E-RUNTIME-STABILIZATION-RELEASE-CLOSE-01-STABLE
CP-40A-POST-RELEASE-SLO-DRIFT-WATCH-01-STABLE

Gobernanza, routing, multi-GPU readiness

CP-GITNEXUS-FIRST-ACTIVATION-01
CP-SLO-ENFORCEMENT-01
CP-VALIDATION-AUTHORITY-01
CP-AUTONOMOUS-OBSERVABILITY-TRIAGE-01
CP-OPERATOR-INTENT-REASONING-01
CP-MULTIGPU-READINESS-01
CP-DYNAMIC-NODE-REGISTRY-01
CP-AI-LAB-MULTI-NODE-ROUTING-01
CP-INTELLIGENT-FALLBACK-ENGINE-01
CP-CAPABILITY-SCHEDULER-01
CP-49A-POOL-ADMIN-API-READONLY-01
CP-DOC-AUTOMATION-STABLE
CP-DOCS-ASTRO-ARCHITECTURE-UPDATE-01-STABLE
CP-DOCS-AILAB-MCP-INFRASTRUCTURE-UPDATE-01-STABLE
CP-MCP-OPENCODE-WINDOWS-CONNECTION-01-STABLE
CP-MCP-SEMANTIC-GATEWAY-01-STABLE

Hermes Enterprise

CP-E01A-SOUL-ENTERPRISE-SKELETON-STABLE
CP-E02A-CAPABILITY-REGISTRY-SKELETON-STABLE
CP-E02B-CAPABILITY-REGISTRY-VALIDATOR-STABLE
CP-E02C-OPERATOR-REGISTRY-VALIDATOR-STABLE
CP-E03A-OPERATOR-REGISTRY-SKELETON-STABLE
CP-E04A-HOOK-REGISTRY-SKELETON-STABLE
CP-E05-MCP-REGISTRY-SKELETON-STABLE
CP-E06-DYNAMIC-GOVERNANCE-STABLE
CP-E07-ENTERPRISE-RUNTIME-STATUS-ENDPOINT-STABLE
CP-HERMES-ENTERPRISE-CORE-01
CP-HERMES-ENTERPRISE-FOUNDATION-01
CP-HERMES-OPERABILITY-TUNING-01
CP-HERMES-DOCS-ASTRO-ENTERPRISE-01

Knowledge Base (AnythingLLM)

CP-ANYTHINGLLM-ENTERPRISE-04-COMPLETE

Límite explícito

Este estado es pre-Multi-GPU. RX7900XT sigue siendo inventario. No hay scheduler Multi-GPU documentado como implementado. Aunque las precondiciones semánticas (30H–31B) están cerradas desde CP-31B, el nodo RX7900XT permanece apagado y no hay plan activo de reactivación.