Unified CLI & Telemetry Suite

The ColabBio Unified Management CLI (cb-svc-cli) provides DevOps, MLOps, pathologists, and bioinformaticians with a sovereign, terminal-native control plane for cluster observation, AI pipelines, model lifecycle, tenant auditing, user management, and conversational clinical assistance.

┌────────────────────────────────────────────────────────────────────────┐
│               COLABBIO UNIFIED MANAGEMENT CLI (cb-svc-cli)             │
├────────────────────────────────────────────────────────────────────────┤
│ • Spring Shell 3.2 + Native Wrapper (`cb-devops/scripts/colabbio`)     │
│ • User RBAC & Clinical Authentication (`colabbio user`)                │
│ • Dynamic Model Loading, Champion/Challenger & MLflow Hub              │
│ • Dual Execution Mode: Connected Live Cluster & Standalone Fallback    │
│ • Interactive AI Companion with ANSI Rich Widget Rendering             │
│ • Real-Time Diagnostics: MicroK8s, MedVFS, Triton, OMERO, OHDSI DQD   │
└────────────────────────────────────────────────────────────────────────┘

---

1. Quick Start & Installation

The CLI is compiled into a self-contained executable Java JAR within cb-svc-parent and accessed via the convenient wrapper script colabbio.

Build & Setup

# 1. Compile and package the CLI suite
cd cb-svc-parent
mvn clean package -DskipTests=false

# 2. (Optional) Add colabbio to your PATH
export PATH="$PATH:$(pwd)/../cb-devops/scripts"

# 3. Verify installation
colabbio help

---

2. Command Architecture & Reference

The CLI is organized into focused sub-command suites designed for hospital cluster administrators, MLOps engineers, and bioinformaticians.

2.1. User & RBAC Governance (colabbio user)

Manage clinical pathologists, resident doctors, study managers, and data stewards per tenant.

# List clinical users registered in a specific hospital tenant
colabbio user list --tenant hospital-la-fe

# Inspect user metadata, assigned roles, and permission scopes
colabbio user get --username dr.garcia

# Provision a new clinical user
colabbio user create --username dr.novo --email novo@hospital.es --password "Secret2026!" --role PATHOLOGIST --tenant hospital-la-fe

# Assign an additional clinical role to an existing user
colabbio user assign-role --username dr.novo --role ROLE_STUDY_MANAGER

# Delete a user
colabbio user delete --username dr.novo

# View the full RBAC roles and permissions catalog
colabbio user roles

---

2.2. Model Lifecycle & Champion/Challenger (colabbio model)

Manage Triton GPU inference serving, hot-loading of models, and A/B version comparison.

# List all active models in Triton Inference Server
colabbio model list

# Inspect model card, tensor shapes, normalization methods, and SHA-256 digest
colabbio model inspect --name abmil-brca-v1

# Dynamically hot-load a new model into Triton GPU and Clinical Visor catalog
colabbio model load --name ki67_cell_density --version 1 --format ONNX --path /data/models/ki67.onnx --tenant hospital-la-fe

# Compare two versions (Champion v1 vs Challenger v2) on a clinical WSI
colabbio model compare --name abmil-brca-v1 --v1 1 --v2 2 --wsi WSI-BRCA-001

# Promote a challenger version to production Champion in study manifest and visor
colabbio model promote --name abmil-brca-v1 --version 2 --study slidelab-abmil-study.yaml

---

2.3. MLflow Model Registry Hub (colabbio mlflow)

Access sovereign models across tenants and global platform foundation models.

# List models available for a hospital tenant (combining Global + Tenant private)
colabbio mlflow list --tenant hospital-la-fe

# List all models in the entire MLflow registry
colabbio mlflow list --tenant all

# Deploy a model from MLflow directly to Triton and activate it in the clinical visor
colabbio mlflow deploy --model abmil-brca-hospital-la-fe --version 2.1.0 --tenant hospital-la-fe

---

2.4. Node & Infrastructure (colabbio node)

Inspect the host machine, GPU accelerators, Tailscale mesh connectivity, and bootstrap initialization logs.

# Check node hardware, available cores, memory, and manifest status
colabbio node status

# Query deep GPU telemetry via nvidia-smi
colabbio node status --gpu

# View bootstrap and initialization logs
colabbio node logs -n 50

Node Manifest Persistence

During cluster initialization via colabbio-node-init.sh, the system persists the node configuration to /etc/colabbio/node-manifest.json and bootstrap logs to /var/log/colabbio/node-init.log.

---

2.5. Cluster Observability (colabbio cluster)

Audit MicroK8s microservices, Ingress endpoints, and MedVFS storage classes.

# Audit all core pods, services, and Ingress routing in the colabbio namespace
colabbio cluster audit

# Stream logs from a specific core service (omero, triton, fhir, ai)
colabbio cluster logs --service triton --lines 100

---

2.6. Study & Pipeline Monitoring (colabbio study)

Track execution progress, step-by-step pipeline stages (SlideLab, PyTorch WSI-MIL training, Triton registration), and generate validation reports.

# Inspect real-time status of a running study pipeline
colabbio study status --run-id run-tcga-brca-001

# Compile and display the Markdown clinical validation report
colabbio study report --run-id run-tcga-brca-001

---

2.7. Pathology Visor & Ingestion (colabbio visor)

Monitor OMERO.server (Blitz), OMERO.web, and ingest Whole Slide Images (SVS, NDPI, MRXS).

# Check health of OMERO.server (port 4064) and DeepZoom rendering bridge
colabbio visor status

# Batch ingest digital pathology slides into OMERO repository
colabbio visor ingest --path /data/wsi_samples/

---

2.8. Triton Inference Benchmarking (colabbio inference)

Benchmark GPU forward-pass latency and attention heatmap computation under load.

# Run latency benchmark with concurrency level 4
colabbio inference benchmark --model abmil-brca-v1 --concurrency 4

---

2.9. Tenant Governance & OHDSI DQD (colabbio tenant)

Manage hospital tenants and execute Data Quality Dashboard (DQD) checks.

# List registered hospital tenants and active dataset quotas
colabbio tenant list

# Run OHDSI Achilles & DQD conformance and plausibility audit
colabbio tenant audit --tenant hospital-la-fe

---

3. Conversational Clinical AI Companion (ask & chat)

The CLI integrates directly with the Clinical AI Companion backend (cb-svc-ai), enabling engineers and bioinformaticians to query datasets, diagnose failures, and inspect pathology slides in natural language.

Single-Shot (ask)

# Query active models in Triton
colabbio ask "cuales son los modelos en triton"

# Analyze a biopsy slide and generate hotspots
colabbio ask "analiza la biopsia de la slide 1092"

# Pipe AI markdown reports directly to disk
colabbio ask "Genera un reporte del paciente PAT-77402" > reporte.md

Interactive Session (chat)

# Start an interactive REPL conversational session
colabbio chat

# Example session:
colabbio> ¿Cuál es el estado del estudio run-tcga-brca-001?
companion> El estudio se encuentra en progreso (Paso 2/4: SlideLab Feature Extraction).
           Se han procesado 78 de 120 portaobjetos digitales (65%).

colabbio> exit

---

4. A2UI Widget Rendering: Web UI vs CLI Terminal

The backend emits structured Markdown with A2UI embedded widgets using the fenced code syntax:

```widget:score_bar
{
  "label": "Probabilidad de Malignidad",
  "score": 0.9412,
  "threshold": 0.50,
  "status": "HIGH_RISK"
}
```

```widget:omero_viewport_action
{
  "imageId": "1092",
  "x": 18450,
  "y": 9200,
  "zoom": 20,
  "label": "Centrar Visor en Hotspot 1"
}
```

Graceful Degradation in Terminal Mode

When rendered in the CLI, the parser transforms JSON blocks into clean, high-contrast ANSI terminal components:

  ┌─ [SCORE BAR: Probabilidad de Malignidad] ─────────────────────────
  │ Valor: 94.12%  [HIGH_RISK]
  │ Barra: [███████████████████░]
  └────────────────────────────────────────────────────────────

  [🔬 OMERO Viewport Action] Centrar Visor en Hotspot 1
     Coordenadas: X: 18450, Y: 9200, Zoom: 20x (Slide ID: 1092)
     Enlace Visor: https://omero.colabbio.local/webclient/?show=image-1092

---

5. Dual Execution Mode Architecture

To ensure operational reliability in disconnected environments (such as isolated edge VMs or air-gapped clinical networks):

1. Connected Live Mode (🟢 Live Cluster): Connects to cb-svc-ai on http://localhost:8080 (or configured COLABBIO_AI_URL) and queries live Kubernetes API, PostgreSQL, and Triton Inference Server.
2. Standalone Fallback Mode (🟡 Standalone): If the backend is offline or undergoing maintenance, the CLI operates locally with built-in diagnostic engines and simulated clinical models.