### Install Dependencies and Start Development Servers

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/README.md

Installs project dependencies, builds shared packages, and starts the backend and frontend development servers.

```bash
# Install dependencies
npm install

# Build shared packages
npm run build -w packages/types
npm run build -w packages/utils

# Start development servers (backend:3000, frontend:5173)
npm run dev
```

--------------------------------

### Start Development Server

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Starts the development server, which automatically creates a Python virtual environment and installs dependencies on the first run.

```bash
# Start development (auto-creates Python venv on first run)
npm run dev
```

--------------------------------

### Production Multi-GPU Setup Example

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/kvcached/configuration.md

Example YAML configuration for a production multi-GPU setup using sardeenz with vLLM. This demonstrates how to configure multiple model instances across different GPUs.

```yaml
kvcached:
  gpu_memory_utilization: 0.95
  preallocate_pages: true
  log_level: WARNING

router:
  enabled: true
  host: '0.0.0.0'
  port: 8080
  timeout: 600

sleep_manager:
  enabled: true
  idle_threshold_seconds: 600 # 10 minutes
  auto_sleep_enabled: true
  min_sleep_duration_seconds: 120
  check_interval_seconds: 120

instances:
  - name: 'llama-70b-gpu0-1'
    model: 'meta-llama/Llama-2-70b-hf'
    engine: 'vllm'
    port: 12346
    env:
      ENABLE_KVCACHED: 'true'
      KVCACHED_AUTOPATCH: '1'
      CUDA_VISIBLE_DEVICES: '0,1'
    engine_args:
      disable_log_requests: true
      enable_prefix_caching: false
      tensor_parallel_size: 2 # Use 2 GPUs

  - name: 'codellama-34b-gpu2'
    model: 'codellama/CodeLlama-34b-hf'
    engine: 'vllm'
    port: 12347
    env:
      ENABLE_KVCACHED: 'true'
      KVCACHED_AUTOPATCH: '1'
      CUDA_VISIBLE_DEVISES: '2'
    engine_args:
      disable_log_requests: true
      enable_prefix_caching: false
      tensor_parallel_size: 1
```

--------------------------------

### Start Multi-Pod Cluster Backend

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/quickstart.md

Configure and start two backend pods for a multi-pod cluster setup. Ensure that CLUSTER_PEERS and CLUSTER_SECRET are consistent across all pods. Each pod simulates 2 GPUs with 24 GB memory.

```bash
# Terminal 1: Pod A
PORT=3001 \
INFERENCE_BACKEND=inference-sim \
DEV_VIRTUAL_GPU_COUNT=2 \
SIM_GPU_MEMORY_GB=24 \
CLUSTER_PEERS=localhost:3001,localhost:3002 \
CLUSTER_SECRET=$(openssl rand -hex 32) \
npm run dev -w apps/backend

# Terminal 2: Pod B (same CLUSTER_SECRET)
PORT=3002 \
INFERENCE_BACKEND=inference-sim \
DEV_VIRTUAL_GPU_COUNT=2 \
SIM_GPU_MEMORY_GB=24 \
CLUSTER_PEERS=localhost:3001,localhost:3002 \
CLUSTER_SECRET=<same-secret-as-above> \
npm run dev -w apps/backend
```

--------------------------------

### Manual Python Environment Setup and Activation

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Manually sets up the Python virtual environment, installs all dependencies including the dev group, and activates the environment. This is an alternative to the automated setup.

```bash
cd apps/backend

# Create venv and install all dependencies (including dev group)
uv sync --locked --group dev

# Activate the environment
source .venv/bin/activate
```

--------------------------------

### Start Development Server

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/apps/backend/CLAUDE.md

Starts the backend development server on port 3000. Use this for local development.

```bash
npm run dev -w apps/backend
```

--------------------------------

### Install Inference-Sim Binary

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/testing.md

Clone the repository, build the binary, and copy it to the system's PATH. Verify the installation by checking the help message.

```bash
git clone https://github.com/llm-d/llm-d-inference-sim.git
cd llm-d-inference-sim
make build
sudo cp bin/llm-d-inference-sim /usr/local/bin/
llm-d-inference-sim --help  # verify
```

--------------------------------

### Complete Model Instance Configuration Example

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/kvcached/configuration.md

A comprehensive example showing the configuration for a model instance, including environment variables and vLLM engine arguments.

```yaml
instances:
  - name: 'llama-3.2-1b-main'
    model: 'meta-llama/Llama-3.2-1B'
    engine: 'vllm'
    port: 12346
    venv: '~/vllm-env'

    env:
      ENABLE_KVCACHED: 'true'
      KVCACHED_AUTOPATCH: '1'
      CUDA_VISIBLE_DEVICES: '0'

    engine_args:
      disable_log_requests: true
      enable_prefix_caching: false
      tensor_parallel_size: 1
      gpu_memory_utilization: 0.9
      max_model_len: 4096
      dtype: 'auto'
```

--------------------------------

### Start Development Servers

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/CLAUDE.md

Starts the backend and frontend development servers for local development. Ensure prerequisites like Node.js and GPU drivers are met.

```bash
npm run dev
```

--------------------------------

### Start Single Pod Backend in Inference-Sim Mode

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/quickstart.md

Launch the backend in inference-sim mode for a single pod setup. This configuration creates a simulated GPU with 24 GB of memory and does not require actual GPU hardware.

```bash
cd apps/backend
INFERENCE_BACKEND=inference-sim npm run dev
```

--------------------------------

### Local Multi-Pod Development Setup

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/004-school-orchestration/quickstart.md

Use these commands to simulate a multi-pod environment locally. Ensure you have Node.js 22 and npm workspaces installed. Run each command in a separate terminal.

```bash
# Terminal 1: Pod A (will become leader)
PORT=3000 CLUSTER_PEERS=localhost:3000,localhost:3001 \
  CLUSTER_SECRET=dev-secret-change-me \
  DEV_VIRTUAL_GPU_COUNT=2 \
  npm run dev -w apps/backend

# Terminal 2: Pod B
PORT=3001 CLUSTER_PEERS=localhost:3000,localhost:3001 \
  CLUSTER_SECRET=dev-secret-change-me \
  DEV_VIRTUAL_GPU_COUNT=2 \
  npm run dev -w apps/backend

# Terminal 3: Frontend (connects to leader)
npm run dev -w apps/frontend
```

--------------------------------

### Start Individual Workspaces

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/README.md

Starts development servers for specific workspaces, such as the backend or frontend.

```bash
# Start individual workspaces
npm run dev -w apps/backend
npm run dev -w apps/frontend
```

--------------------------------

### Install and Serve Controller API with Swagger UI

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/contracts/README.md

Install the swagger-ui-watcher tool globally and then use it to serve the controller API specification locally.

```bash
npm install -g swagger-ui-watcher

swagger-ui-watcher controller-api.yaml
```

--------------------------------

### Install React Data View

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/components/data-display/README.md

Install the @patternfly/react-data-view package using npm.

```bash
npm install @patternfly/react-data-view
```

--------------------------------

### Verify Inference Simulator Installation

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Check if the inference simulator has been installed correctly by running its help command.

```bash
llm-d-inference-sim --help
```

--------------------------------

### Install and Serve Proxy API with Swagger UI

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/contracts/README.md

Install the swagger-ui-watcher tool globally and then use it to serve the proxy API specification locally.

```bash
npm install -g swagger-ui-watcher

swagger-ui-watcher proxy-api.yaml
```

--------------------------------

### Start Backend with Virtual GPUs

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Use this command to start the backend with a specified number of virtual GPUs. This is useful for testing multi-GPU features when you only have a single physical GPU.

```bash
# Start backend with 2 virtual GPUs
DEV_VIRTUAL_GPU_COUNT=2 npm run dev -w apps/backend

# In another terminal
npm run dev -w apps/frontend
```

--------------------------------

### Sardeenz Configuration File Example

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/deployment.md

Example configuration for the Sardeenz application, covering server, authentication, GPU, model, logging, and metrics settings.

```yaml
server:
  port: 3000 # Unified API port
  routingOverheadTarget: 50 # ms (p95) for proxy routing

auth:
  enabled: true
  provider: oauth
  issuerUrl: https://oauth-openshift.apps.your-cluster.com
  k8sApiUrl: https://api.your-cluster.com:6443
  clientId: sardeenz

gpu:
  memoryReserve: 2.0 # GB
  maxModels: 5

models:
  path: /models
  preloadOnStartup: [] # Model IDs to preload

logging:
  level: info
  format: json

metrics:
  enabled: true
  port: 9090

```

--------------------------------

### PatternFly React Repository Structure Example

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/resources/external-links.md

Illustrates the directory structure within the patternfly-react repository, highlighting locations for component examples and specialized packages. Useful for locating source code and usage examples.

```bash
patternfly-react/
├── packages/react-core/src/components/
│   ├── Button/examples/
│   ├── Table/examples/
│   ├── Dropdown/examples/
│   └── [ComponentName]/examples/
├── packages/react-table/src/components/
└── packages/react-charts/src/components/
```

--------------------------------

### Install PatternFly Component Groups

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/component-groups/README.md

Install the @patternfly/react-component-groups package using npm.

```bash
npm install @patternfly/react-component-groups
```

--------------------------------

### Install All Workspace Dependencies

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/quickstart.md

Install all project dependencies for the root workspace and its sub-workspaces (backend, frontend).

```bash
npm install
```

--------------------------------

### Start Single-Pod Backend

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/testing.md

Run the backend in single-pod mode using npm. This command starts the development server for the backend application.

```bash
INFERENCE_BACKEND=inference-sim npm run dev -w apps/backend
```

--------------------------------

### Check Node.js and npm Installation

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/troubleshooting/common-issues.md

Verify if Node.js and npm are installed and accessible in your system's PATH.

```bash
# Check if Node.js is installed
node --version
npm --version
```

--------------------------------

### Start Multi-Pod Cluster

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/testing.md

Set up and start a multi-pod cluster for the inference backend. This involves setting environment variables for port, GPU count, memory, and cluster peers.

```bash
export SECRET=$(openssl rand -hex 32)

# Terminal 1
PORT=3001 INFERENCE_BACKEND=inference-sim DEV_VIRTUAL_GPU_COUNT=2 \
  SIM_GPU_MEMORY_GB=24 CLUSTER_PEERS=localhost:3001,localhost:3002 \
  CLUSTER_SECRET=$SECRET npm run dev -w apps/backend

# Terminal 2
PORT=3002 INFERENCE_BACKEND=inference-sim DEV_VIRTUAL_GPU_COUNT=2 \
  SIM_GPU_MEMORY_GB=24 CLUSTER_PEERS=localhost:3001,localhost:3002 \
  CLUSTER_SECRET=$SECRET npm run dev -w apps/backend
```

--------------------------------

### Install Frontend Dependencies

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/setup/quick-start.md

Install all project dependencies or only those for the frontend. This includes PatternFly 6, React, Vite, TypeScript, and ESLint.

```bash
# From project root, install all dependencies
npm install

# Or install only frontend dependencies
cd frontend && npm install
```

--------------------------------

### Start Frontend-Only Development

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Starts the frontend development server when only the frontend is needed, without any backend services.

```bash
npm run dev -w apps/frontend
```

--------------------------------

### Start Frontend Development Server

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/setup/development-environment.md

Use these commands to start the Vite development server. You can run them from the project root or navigate to the frontend directory first.

```bash
# Start frontend development server
npm run dev:frontend

# Or from frontend directory
cd frontend && npm run dev

# Stop server
Ctrl+C
```

--------------------------------

### Install PatternFly Packages

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/troubleshooting/common-issues.md

Install necessary PatternFly React packages in your frontend project. Navigate to the 'frontend' directory before running this command.

```bash
cd frontend
npm install @patternfly/react-core @patternfly/react-table @patternfly/react-icons
```

--------------------------------

### Get Benchmark Details Response

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

Example JSON response for a successful GET request to retrieve benchmark details.

```json
{
  "benchmark": {
    "id": "550e8400-e29b-41d4-a716-446655440000",
    "name": "SmolLM Performance Test",
    "status": "completed",
    "mode": "isolated",
    "kvcached_enabled": true,
    "created_at": "2025-11-29T12:00:00Z",
    "completed_at": "2025-11-29T12:02:30Z",
    "total_requests": 20,
    "successful_requests": 20,
    "failed_requests": 0,
    "duration_seconds": 149.5,
    "scenarios": [
      {
        "id": "scenario-uuid",
        "instance_id": "smollm2-135m-abc123",
        "routing_mode": "direct",
        "model_path": "HuggingFaceTB/SmolLM2-135M-Instruct",
        "model_name": "SmolLM2-135M-Instruct",
        "input_tokens": 100,
        "output_tokens": 50,
        "concurrency": 4,
        "status": "completed",
        "metrics": {
          "ttft_avg": 45.2,
          "ttft_p50": 42.1,
          "ttft_p90": 68.3,
          "ttft_p95": 78.5,
          "ttft_p99": 95.2,
          "tps_avg": 156.8,
          "tps_p50": 158.2,
          "tps_p90": 142.3,
          "tps_p95": 138.5,
          "e2e_avg": 320.5,
          "e2e_p50": 315.2,
          "e2e_p90": 385.3,
          "e2e_p95": 412.1,
          "e2e_p99": 478.5,
          "goodput_count": 18,
          "goodput_percent": 90.0,
          "requests_per_second": 12.5,
          "tokens_per_second_total": 1960.0,
          "total_requests": 20,
          "successful_requests": 20,
          "failed_requests": 0
        }
      }
    ]
  }
}
```

--------------------------------

### Install Kubernetes Client Node Dependency

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/004-school-orchestration/tasks.md

Installs the @kubernetes/client-node package for the backend application. This is a setup task for cluster support.

```bash
npm install @kubernetes/client-node -w apps/backend
```

--------------------------------

### Parallel Package and Plugin Tasks (Foundational Phase)

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/tasks.md

Demonstrates how multiple package initialization tasks and backend plugin creation tasks can be executed concurrently during the foundational phase.

```bash
# All shared package tasks can run together:
Task T008: "Initialize packages/types workspace"
Task T009: "Initialize packages/utils workspace"
Task T010: "Initialize packages/contracts workspace"

# All backend plugins can run together:
Task T029: "Create OAuth 2.0 plugin in apps/backend/src/plugins/auth.ts"
Task T030: "Create Swagger plugin in apps/backend/src/plugins/swagger.ts"
Task T031: "Create Prometheus metrics plugin in apps/backend/src/plugins/metrics.ts"
```

--------------------------------

### Install Node.js Dependencies and Build Packages

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Installs Node.js dependencies and builds specific packages within the project. This is part of the initial setup for the development environment.

```bash
# Install Node.js dependencies
npm install
npm run build -w packages/types && npm run build -w packages/utils
```

--------------------------------

### Get Model Details Request

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

Example cURL command to fetch details for a specific model instance.

```bash
curl -H "Authorization: Bearer $TOKEN" \
  http://localhost:3000/api/v1/models/llama-3-2-1b-abc123
```

--------------------------------

### Run Development with Custom Simulator Configuration

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Start the development server with custom inference simulator settings, such as specifying the number of GPUs and their memory.

```bash
# Single GPU with 48 GB
INFERENCE_BACKEND=inference-sim DEV_VIRTUAL_GPU_COUNT=1 SIM_GPU_MEMORY_GB=48 npm run dev
```

--------------------------------

### List Memory Profiles Response

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

Example JSON response for a successful GET request to list memory profiles.

```json
{
  "profiles": [
    {
      "id": "profile-uuid",
      "profile_name": "SmolLM2-135M @ 2048 tokens",
      "model_path": "HuggingFaceTB/SmolLM2-135M-Instruct",
      "max_tokens": 2048,
      "total_gpu_memory_gib": 1.22,
      "weights_memory_gib": 0.27,
      "cuda_graphs_gib": 0.55,
      "overhead_memory_gib": 0.4,
      "kv_cache_available_gib": 5.7,
      "gpu_name": "NVIDIA GeForce RTX 4090",
      "gpu_total_memory_gib": 24.0,
      "created_at": "2025-11-29T12:00:00Z"
    }
  ],
  "total": 1
}
```

--------------------------------

### Complete kvctl Session Example

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/kvcached/cli-tools.md

Demonstrates a typical session using kvctl, including launching the tool, listing IPC segments, setting memory limits, verifying limits, and exiting.

```bash
# Launch kvctl
$ kvctl

# List current IPC segments
kvctl> list
IPC Name         Total      Used       Free       Limit      Usage
─────────────────────────────────────────────────────────────────
VLLM_MODEL_1     8.0 GB     0.0 GB     8.0 GB     None       0%
VLLM_MODEL_2     6.0 GB     0.0 GB     6.0 GB     None       0%

# Set memory limits
kvctl> limit VLLM_MODEL_1 6G
Memory limit for VLLM_MODEL_1 set to 6.0 GB

kvctl> limit-percent VLLM_MODEL_2 25
Memory limit for VLLM_MODEL_2 set to 25% (6.0 GB of 24 GB total)

# Verify limits
kvctl> list
IPC Name         Total      Used       Free       Limit      Usage
─────────────────────────────────────────────────────────────────
VLLM_MODEL_1     8.0 GB     0.0 GB     8.0 GB     6.0 GB     0%
VLLM_MODEL_2     6.0 GB     0.0 GB     6.0 GB     6.0 GB     0%

# Watch memory usage in real-time
kvctl> watch -n 2
(Refreshes every 2 seconds...)

# Exit
kvctl> exit
```

--------------------------------

### Build Inference Simulator from Source

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Build the inference simulator binary from source if you have Go 1.22+ installed. This binary mimics vLLM without requiring GPU hardware.

```bash
git clone https://github.com/llm-d/llm-d-inference-sim.git
cd llm-d-inference-sim
make build
sudo cp bin/llm-d-inference-sim /usr/local/bin/
```

--------------------------------

### Get Model Instance Logs Request

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

Example cURL command to fetch the last 50 log lines for a model instance.

```bash
curl -H "Authorization: Bearer $TOKEN" \
  "http://localhost:3000/api/v1/models/instances/abc123/logs?lines=50"
```

--------------------------------

### Start Backend in Production Mode

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/quickstart.md

Run the backend application in production mode.

```bash
# Start backend in production mode
npm run start -w apps/backend
```

--------------------------------

### vLLM Health Check Endpoint

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/architecture.md

Basic HTTP GET request to check the health status of a vLLM instance. Assumes a local setup.

```bash
GET http://localhost:{port}/health
```

--------------------------------

### Get System Metrics

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

Retrieve system-wide metrics in Prometheus format, including the total number of active, starting, stopping, and failed model instances.

```prometheus
# HELP sardeenz_models_total Total number of model instances
# TYPE sardeenz_models_total gauge
sardeenz_models_total{status="active"} 2
sardeenz_models_total{status="starting"} 0
sardeenz_models_total{status="stopping"} 0
sardeenz_models_total{status="failed"} 0

# HELP sardeenz_gpu_memory_used_bytes GPU memory used by all models

```

--------------------------------

### Get Memory Usage

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/testing.md

Retrieve the current memory usage of the inference backend. Two endpoints are available: one for general usage and another for multi-GPU setups.

```bash
curl http://localhost:3000/api/memory/usage
```

```bash
curl http://localhost:3000/api/memory/usage/multi-gpu
```

--------------------------------

### Fetch Memory Usage with curl

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

This command-line example shows how to make a GET request to the memory usage endpoint using curl. Ensure you replace $TOKEN with your actual authentication token.

```bash
curl -H "Authorization: Bearer $TOKEN" \
  http://localhost:3000/api/memory/usage
```

--------------------------------

### Deploy to Production Environment

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/deployment/README.md

Apply the Kustomize configuration for the production environment to deploy the Sardeenz application.

```bash
oc apply -k deployment/overlays/prod/
```

--------------------------------

### Set Up DndContext and Drag Overlay

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/003-model-move/plan.md

Wrap the application content with `DndContext` to enable drag-and-drop functionality. Use `DragOverlay` to render a visual representation of the item being dragged. This snippet shows the basic setup and event handlers for drag start and end.

```typescript
import { DndContext, DragOverlay, pointerWithin } from '@dnd-kit/core'

// Wrap content in DndContext
<DndContext
  collisionDetection={pointerWithin}
  onDragStart={handleDragStart}
  onDragEnd={handleDragEnd}
>
  {/* GPU groups */}
  <DragOverlay>
    {activeDragModel && <ModelCardCompact model={activeDragModel} isDragOverlay />}
  </DragOverlay>
</DndContext>

// Handle drop
function handleDragEnd(event: DragEndEvent) {
  const { active, over } = event
  if (!over) return

  const model = active.data.current?.model
  const targetGpuIds = over.data.current?.gpuIds

  if (model && targetGpuIds) {
    // Open move dialog with preselected target
    setMoveDialogModel(model)
    setPreselectedGpuIds(targetGpuIds)
    setIsMoveDialogOpen(true)
  }
}
```

--------------------------------

### Pod Initialization and Configuration

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/architecture-visualizer.html

Details the initial startup of a Sardeenz pod, including environment variables for pod name and cluster configuration.

```text
POD_NAME=sardeenz-0
CLUSTER_SECRET=<32+ char secret>
CLUSTER_EXPECTED_PODS=2
```

--------------------------------

### Complete Workflow Example in TypeScript

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

This TypeScript code demonstrates a full workflow using the Controller and Proxy APIs. It covers loading a model, waiting for it to become ready, making an inference request, and finally unloading the model. Ensure you have axios installed and the AUTH_TOKEN environment variable set.

```typescript
import axios from 'axios'

const CONTROLLER_URL = 'http://localhost:3000/api/v1'
const PROXY_URL = 'http://localhost:8000/v1'
const AUTH_TOKEN = process.env.AUTH_TOKEN

async function completeWorkflow() {
  // 1. Load a model
  console.log('Loading model...')
  const loadResponse = await axios.post(
    `${CONTROLLER_URL}/models/load`,
    {
      modelPath: '/models/meta-llama/Llama-3.2-1B',
      displayName: 'Llama 3.2 1B',
      gpuMemoryLimit: 4.0,
      port: 5001,
    },
    {
      headers: { Authorization: `Bearer ${AUTH_TOKEN}` },
    }
  )

  const modelId = loadResponse.data.id
  const instanceId = loadResponse.data.instance_id
  console.log(`Model loading started: ${instanceId}`)

  // 2. Wait for model to be ready
  // Option A: Polling (shown below)
  // Option B: Subscribe to SSE at /api/v1/models/instances/{instance_id}/events
  console.log('Waiting for model to be ready...')
  let status = 'starting'
  while (status === 'starting') {
    await new Promise((resolve) => setTimeout(resolve, 5000))
    const statusResponse = await axios.get(`${CONTROLLER_URL}/models/${modelId}`, {
      headers: { Authorization: `Bearer ${AUTH_TOKEN}` },
    })
    status = statusResponse.data.status
    console.log(`Status: ${status}`)
  }

  if (status !== 'active') {
    throw new Error(`Model failed to start: ${status}`)
  }

  // 3. Make inference request
  console.log('Making inference request...')
  const inferenceResponse = await axios.post(`${PROXY_URL}/chat/completions`, {
    model: modelId,
    messages: [{ role: 'user', content: 'What is 2+2?' }],
  })

  console.log('Response:', inferenceResponse.data.choices[0].message.content)

  // 4. Unload model
  console.log('Unloading model...')
  await axios.post(
    `${CONTROLLER_URL}/models/${modelId}/unload`,
    {},
    { headers: { Authorization: `Bearer ${AUTH_TOKEN}` } }
  )

  console.log('Workflow complete!')
}

completeWorkflow().catch(console.error)
```

--------------------------------

### Install PatternFly Codemods Globally

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/guidelines/migration-codemods.md

Installs the `@patternfly/pf-codemods` package globally after clearing the npm cache to ensure a clean installation.

```bash
# Clear npm cache
npm cache clean --force

# Install globally
npm install -g @patternfly/pf-codemods
```

--------------------------------

### Parallel Tasks Example: Phase 1

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/tasks.md

Illustrates tasks that can run in parallel after foundational configuration is complete. These tasks involve creating utility files and their corresponding unit tests.

```bash
# After T001 (config) completes, launch all 4 tasks in parallel:
Task T002: "Create ModelMemoryEstimator in apps/backend/src/utils/model-memory-estimator.ts"
Task T003: "Create SimGpuTracker in apps/backend/src/utils/sim-gpu-tracker.ts"
Task T004: "Unit tests for ModelMemoryEstimator in apps/backend/tests/unit/model-memory-estimator.test.ts"
Task T005: "Unit tests for SimGpuTracker in apps/backend/tests/unit/sim-gpu-tracker.test.ts"
```

--------------------------------

### Install uv package manager

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/dev-setup.md

Installs the uv package manager using a script. Ensure to source your bashrc or restart your terminal after installation.

```bash
curl -LsSf https://astral.sh/uv/install.sh | sh
source ~/.bashrc  # or restart terminal
```

--------------------------------

### kvtop Display Layout Example

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/kvcached/cli-tools.md

Illustrates the typical display layout of the kvtop tool, showing overall GPU memory usage and individual IPC segment details with memory bars.

```text
┌─────────────────────────────────────────────────────────────────┐
│                         kvcached Memory Monitor                  │
│                      Refresh: 1.0s | Press q to quit             │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│  GPU Memory: 24.0 GB Total | 15.2 GB Used | 8.8 GB Free (63%)   │
│  ████████████████████████████████████░░░░░░░░░░░░░░░░░░         │
│                                                                  │
├─────────────────────────────────────────────────────────────────┤
│  IPC Segment: VLLM_MODEL_1                                       │
│  Memory: 8.0 GB Total | 4.5 GB Used | 3.5 GB Free (56%)          │
│  Limit: 6.0 GB                                                   │
│  ████████████████████████████████████░░░░░░░░░░░░░░░            │
│                                                                  │
│  IPC Segment: VLLM_MODEL_2                                       │
│  Memory: 6.0 GB Total | 2.1 GB Used | 3.9 GB Free (35%)          │
│  Limit: 6.0 GB                                                   │
│  ████████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            │
│                                                                  │
│  IPC Segment: VLLM_MODEL_3                                       │
│  Memory: 4.0 GB Total | 0.8 GB Used | 3.2 GB Free (20%)          │
│  Limit: None                                                     │
│  ████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░            │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘
```

--------------------------------

### Install PatternFly Charts Dependencies

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/troubleshooting/common-issues.md

Install the necessary dependencies for using PatternFly charts. This command installs both the core charts library and the Victory charting engine.

```bash
npm install @patternfly/react-charts victory
```

--------------------------------

### Configure Backend Environment Variables

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/quickstart.md

Create the .env file for the backend service, setting essential configurations for the server, OAuth, vLLM, and logging.

```bash
# Create environment file for backend
cat > apps/backend/.env <<EOF
# Server Configuration
PORT=3000
NODE_ENV=development

# OAuth 2.0 Configuration
OAUTH_CLIENT_ID=sardeenz
OAUTH_CLIENT_SECRET=change-me-in-production
OAUTH_ISSUER_URL=https://oauth-openshift.apps.example.com
K8S_API_URL=https://api.example.com:6443
JWT_SECRET=change-me-in-production
API_BASE_URL=http://localhost:3000

# vLLM Configuration
VLLM_BASE_PORT=12346
VLLM_MAX_INSTANCES=10

# kvcached Configuration
ENABLE_KVCACHED=true
KVCACHED_AUTOPATCH=1

# Logging
LOG_LEVEL=info
EOF
```

--------------------------------

### Install CUDA Toolkit 12.4 on Ubuntu 22.04

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/quickstart.md

Installs the CUDA Toolkit version 12.4 on Ubuntu 22.04 systems. This involves downloading a key, updating package lists, and installing the toolkit.

```bash
# Install CUDA Toolkit (if not already installed)
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt-get install -y cuda-toolkit-12-4

# Verify CUDA installation
nvidia-smi  # Should show CUDA version 12.x
```

--------------------------------

### Frontend Build and Preview Commands

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/setup/development-environment.md

Commands to build the frontend application for production and preview the production build locally. Ensure you are in the correct directory before running.

```bash
# Build frontend for production
npm run build:frontend

# Preview production build
cd frontend && npm run preview
```

--------------------------------

### Install Victory and ECharts

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/charts/README.md

Install the necessary packages for Victory-based or ECharts-based charts. Victory is recommended.

```bash
# ✅ Victory-based charts (recommended)
npm install @patternfly/react-charts victory

# ✅ ECharts-based charts (alternative)
npm install @patternfly/react-charts echarts
```

--------------------------------

### Serve Frontend Production Build

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/quickstart.md

Serve the production build of the frontend application using the 'serve' package.

```bash
# Serve frontend build (requires serve package)
npx serve apps/frontend/dist
```

--------------------------------

### Install PatternFly 6 Packages

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/research.md

Installs the core, table, and icons packages for PatternFly version 6.

```bash
cd apps/frontend
npm install @patternfly/react-core@6 @patternfly/react-table@6 @patternfly/react-icons@6
```

--------------------------------

### Deploy Using Kustomize

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/deployment/README.md

Applies all Sardeenz resources using Kustomize. Ensure you edit kustomization.yaml to set your image registry before applying.

```bash
# Edit kustomization.yaml to set your image registry
# Then apply all resources
oc apply -k deployment/
# or
kubectl apply -k deployment/
```

--------------------------------

### Import Benchmarks

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/api-guide.md

Import benchmark runs from a backup file. Runs that already exist will be skipped.

```bash
curl -X POST http://localhost:3000/api/cluster/benchmarks/import \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d @benchmarks-backup.json
```

--------------------------------

### KVCACHED_IPC_NAME Examples

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/kvcached/configuration.md

Examples of KVCACHED_IPC_NAME settings for different GPU configurations, automatically set by the sardeenz backend.

```bash
# Single GPU on GPU 0
export KVCACHED_IPC_NAME="kvcached_vllm_GPU0"

# Single GPU on GPU 1
export KVCACHED_IPC_NAME="kvcached_vllm_GPU1"

# Tensor-parallel on GPUs 0 and 1
export KVCACHED_IPC_NAME="kvcached_vllm_GPU0_GPU1"

# Tensor-parallel on GPUs 0-3
export KVCACHED_IPC_NAME="kvcached_vllm_GPU0_GPU1_GPU2_GPU3"
```

--------------------------------

### Verify vLLM Installation

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/quickstart.md

Checks if the vLLM binary is installed and accessible in the system's PATH, and displays its version.

```bash
# Verify vLLM installation
which vllm
vllm --version
```

--------------------------------

### Install with Legacy Peer Dependencies

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/pf6-guide/troubleshooting/common-issues.md

Use the '--legacy-peer-deps' flag during 'npm install' to bypass peer dependency conflicts.

```bash
npm install --legacy-peer-deps
```

--------------------------------

### PatternFly Setup in main.tsx

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/research.md

Initializes the React application, sets up routing with BrowserRouter, and imports base PatternFly CSS.

```typescript
// apps/frontend/src/main.tsx
import React from 'react'
import ReactDOM from 'react-dom/client'
import { BrowserRouter } from 'react-router-dom'
import '@patternfly/react-core/dist/styles/base.css'
import App from './App'

ReactDOM.createRoot(document.getElementById('root')!).render(
  <React.StrictMode>
    <BrowserRouter>
      <App />
    </BrowserRouter>
  </React.StrictMode>
)
```

--------------------------------

### ModelInstance State Transitions

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/001-multi-model-platform/data-model.md

Illustrates the lifecycle state transitions for a ModelInstance, from starting to active, stopping, and potential failure.

```text
┌─────────┐
│ starting│──┐
└─────────┘  │
             │ Health check succeeds
             ▼
         ┌────────┐
         │ active │
         └────────┘
             │
             │ Unload requested
             ▼
         ┌──────────┐
         │ stopping │
         └──────────┘
             │
             │ Process exits
             ▼
         (deleted)

         ┌─────────┐
         │ starting│
         └─────────┘
             │
             │ Health check fails or process crashes
             ▼
         ┌────────┐
         │ failed │
         └────────┘
```

--------------------------------

### Metrics API: Get All Metrics Function

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/frontend-api-client.md

Retrieves resource metrics for all models by sending a GET request to the '/metrics' endpoint.

```tsx
getAllMetrics: () => apiClient.get<ResourceMetrics[]>('/metrics'),
```

--------------------------------

### Metrics API: Get System Metrics Function

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/docs/development/frontend-api-client.md

Retrieves system-wide metrics by sending a GET request to the '/metrics/system' endpoint.

```tsx
getSystemMetrics: () => apiClient.get<SystemMetrics>('/metrics/system'),
```

--------------------------------

### Project Structure: Documentation

Source: https://github.com/rh-aiservices-bu/sardeenz/blob/main/specs/006-inference-sim-backend/plan.md

Overview of the documentation files for the inference simulator backend feature.

```text
specs/006-inference-sim-backend/
├── plan.md              # This file
├── research.md          # Phase 0 output
├── data-model.md        # Phase 1 output
├── quickstart.md        # Phase 1 output
└── tasks.md             # Phase 2 output (via /speckit.tasks)
```