Setup Scripts

PowerShell scripts for automated setup and management.

Overview

Script	Purpose
setup-ollama.ps1	Main Ollama installation and model setup
setup-ollama-websearch.ps1	Web search integration (Open WebUI, Perplexica)
optimize-ollama-5090.ps1	Ensure models run 100% on GPU, cleanup duplicates
update-pal-config.ps1	Sync PAL MCP config with Ollama models
setup-uncensored-models.ps1	Download uncensored/unfiltered models
limit-ollama-bandwidth.ps1	Limit download bandwidth (requires Admin)
scan-ollama-models.ps1	Scan Ollama library for new coder models
move-ollama-models.ps1	Move models to another drive (requires Admin)
backup-ollama-models.ps1	Backup/restore models to external storage
debug-ollama-connection.ps1	Diagnose container connectivity issues
test-ollama-stack.ps1	Test suite for verifying setup
test-searxng-engines.ps1	Test each SearXNG search engine individually
cleanup-containers.ps1	Stop and remove stack containers
sync-container-images.ps1	Mirror container images to your registry

setup-ollama.ps1

The main setup script for Ollama and models.

Usage

# Default: Install Ollama + recommended models
.\setup-ollama.ps1

# Minimal models only (qwen2.5-coder:32b)
.\setup-ollama.ps1 -MinimalModels

# All recommended models (including community finetunes)
.\setup-ollama.ps1 -AllModels

# Coding-focused models only (10 models)
.\setup-ollama.ps1 -CoderModels

# Skip model downloads
.\setup-ollama.ps1 -SkipModels

# List available models without installing
.\setup-ollama.ps1 -ListModels

Coder Models (-CoderModels)

The -CoderModels flag installs 10 coding-focused models:

Model	Size	Description
qwen2.5-coder:32b	~19GB	Best local coding model
qwen3-coder:30b	~19GB	256K context MoE
devstral-small-2	~15GB	384K context, 65.8% SWE-Bench
devstral	~14GB	Agentic coding
mikepfunk28/deepseekq3_coder	~5GB	DeepSeek + Qwen3 thinking
mikepfunk28/deepseekq3_agent	~5GB	Agent-focused with tools
second_constantine/deepseek-coder-v2:16b	~9GB	160K MoE coder
qwen2.5-coder:14b	~9GB	Efficient coding
deepseek-coder:33b	~19GB	87 language support
codellama:34b	~20GB	Meta's coding model

What It Does

1. Checks prerequisites - NVIDIA drivers, GPU detection
2. Installs Ollama - Downloads and installs if not present
3. Configures environment - Sets OLLAMA_HOST for container access
4. Downloads models - Smart detection (only downloads missing models)
5. Updates MCP config - Configures custom_models.json for PAL

Model Selection

Models are selected based on your VRAM:

VRAM	Models
32GB+	qwen3:32b, deepseek-r1:32b, llama3.3:70b-q4
24GB	qwen3:32b, deepseek-r1:32b, llama3.1:8b
16GB	qwen3:14b, llama3.1:8b, mistral:7b
12GB	llama3.1:8b, mistral:7b, phi-4:14b-q4

setup-ollama-websearch.ps1

Installs web search integration with Open WebUI or Perplexica. SearXNG is always bundled for multi-engine web search.

Usage

# Recommended: Single-user mode (no login required)
.\setup-ollama-websearch.ps1 -Setup OpenWebUI -SingleUser

# Interactive menu (prompts for choice)
.\setup-ollama-websearch.ps1

# Install Perplexica + SearXNG
.\setup-ollama-websearch.ps1 -Setup Perplexica

# Install both
.\setup-ollama-websearch.ps1 -Setup Both

# Skip model downloads
.\setup-ollama-websearch.ps1 -Setup OpenWebUI -SkipModels

# Use mirrored images from your registry
.\setup-ollama-websearch.ps1 -UseLocalRegistry

# Run tests on existing setup
.\setup-ollama-websearch.ps1 -Test

# Troubleshoot connectivity issues
.\setup-ollama-websearch.ps1 -Diagnose

# Remove all web search containers
.\setup-ollama-websearch.ps1 -Uninstall

Models Installed (RTX 5090 Optimized)

The script automatically pulls models optimized for 32GB VRAM:

Model	VRAM	Purpose
qwen2.5:3b	~4GB	Fast web search queries
qwen2.5-coder:14b	~17GB	Synthesis and code

Total: ~21GB - both models fit in VRAM simultaneously, leaving ~11GB for context.

Testing (Optional)

By default, the script just sets up containers quickly without waiting for health checks. Use -Test to verify your setup:

.\setup-ollama-websearch.ps1 -Test

The test runs:

Test	Description
Container health	Checks all containers are running
Model inference	Each model responds to a prompt
Web endpoints	Verifies Open WebUI, SearXNG, Perplexica respond

Troubleshooting

Use -Diagnose to troubleshoot connectivity issues:

.\setup-ollama-websearch.ps1 -Diagnose

Shows: container runtime, Ollama status, container status, network config, and logs for unhealthy containers.

Container Runtime Detection

Automatically detects Docker or Podman:

# Preferred order:
# 1. Docker (if running)
# 2. Podman (if running)

Ports Used

Service	Port	Notes
Open WebUI	3000	Chat interface
SearXNG	4000	Always started with Open WebUI
Perplexica Frontend	3002	(if installed)
Perplexica Backend	3001	(if installed)

optimize-ollama-5090.ps1

Ensures Ollama models run 100% on GPU with 0% CPU offloading. Optimized for RTX 5090 (32GB VRAM). Also includes cleanup to remove duplicate models.

The Problem

Large models with default context windows overflow VRAM, causing CPU offloading:

ollama ps
NAME              ID            SIZE     PROCESSOR        UNTIL
deepseek-r1:32b   6e4c38e2f...  67 GB    61%/39% CPU/GPU  4 minutes from now

The "61%/39% CPU/GPU" means 61% of model layers are on CPU - slow!

Usage

# Optimize all installed models
.\optimize-ollama-5090.ps1

# Optimize a specific model
.\optimize-ollama-5090.ps1 -Model "deepseek-r1:32b"

# Show optimization status
.\optimize-ollama-5090.ps1 -List

# Remove all -5090 variants
.\optimize-ollama-5090.ps1 -Undo

# Override context size
.\optimize-ollama-5090.ps1 -ContextSize 32768

# Force re-optimize even if variant exists
.\optimize-ollama-5090.ps1 -Force

# Delete original models where -5090 variant exists (saves 100GB+)
.\optimize-ollama-5090.ps1 -Cleanup

# Auto-delete without prompting
.\optimize-ollama-5090.ps1 -Cleanup -NoPrompt

# Delete original immediately after each optimization
.\optimize-ollama-5090.ps1 -DeleteOriginal

Parameters

Parameter	Description
-Model	Optimize a specific model instead of all
-List	Show optimization status without changes
-Undo	Remove all -5090 variants
-Cleanup	Delete originals where -5090 exists
-DeleteOriginal	Delete original after each successful optimization
-NoPrompt	Skip confirmation prompts (auto-yes)
-ContextSize	Override auto-calculated context size
-Force	Re-optimize even if variant exists

How It Works

1. Test Each Model: Loads model with a simple prompt, checks ollama ps
2. Detect CPU Usage: Parses processor column for CPU/GPU split
3. Create Variant: If CPU > 0%, creates <model>-5090 variant with:
   • num_gpu 99 - force all layers to GPU
   • Calculated num_ctx - reduced context to fit in VRAM
4. Verify: Re-tests the new variant to confirm 100% GPU
5. Track: Saves results to 5090-optimized.json

Context Size by Model

Model Size	num_ctx	Reasoning
< 10GB	65536	Plenty of headroom
10-15GB	32768	Good balance
15-20GB	16384	Tight fit, prioritize GPU
> 20GB	8192	Minimal context, full GPU

Example Output

============================================================
  Ollama GPU Optimizer for RTX 5090 (32GB VRAM)
============================================================

[INFO] Detected GPU with 32GB VRAM
[INFO] Found 15 model(s) to test

Testing deepseek-r1:32b (19GB)...
  Loading model...
  CPU: 61% / GPU: 39% - NEEDS OPTIMIZATION
  Creating deepseek-r1:32b-5090 with num_ctx=16384, num_gpu=99
  Re-test: CPU: 0% / GPU: 100% - OK

Testing qwen3:32b (20GB)...
  Loading model...
  CPU: 0% / GPU: 100% - ALREADY OPTIMIZED

============================================================
  Summary
============================================================

  Tested:            15 model(s)
  Newly optimized:   8
  Already optimized: 7

[OK] Created 8 optimized variant(s)

Usage:
  ollama run <model>-5090
  Example: ollama run deepseek-r1:32b-5090

Using Optimized Models

After optimization, use the -5090 variants:

# Instead of:
ollama run deepseek-r1:32b

# Use:
ollama run deepseek-r1:32b-5090

Recommended

Run this script after installing new models to ensure they're optimized for your GPU.

Cleanup Duplicate Models

After optimization, you have both the original and the -5090 variant. The cleanup feature helps you reclaim storage:

# See what can be cleaned up
.\optimize-ollama-5090.ps1 -List

# Delete originals where -5090 exists
.\optimize-ollama-5090.ps1 -Cleanup

Example output:

Found 10 base model(s) with existing -5090 variants:

  qwen3-coder:30b                             18 GB
  nemotron-3-nano:30b                         24 GB
  deepseek-r1:32b                             19 GB
  qwen2.5-coder:32b                           19 GB
  ...

Total space to free: 147.4 GB

Delete these 10 original model(s)? [y/N]

warning

Once deleted, originals cannot be recovered. You'll need to re-download them if needed. The -5090 variants work identically but with optimized settings.

update-pal-config.ps1

Synchronizes your PAL MCP Server configuration with locally installed Ollama models. PAL is an AI orchestration tool that provides advanced capabilities like multi-model consensus, deep thinking, and systematic code review.

Why Use This Script

When you install or optimize Ollama models, PAL doesn't automatically know about them. This script:

1. Discovers your models - Scans ollama list for all installed models
2. Sets capabilities - Marks reasoning models with supports_extended_thinking
3. Ranks models - Assigns intelligence_score for model selection
4. Creates aliases - Short names like r1 → deepseek-r1:32b-5090
5. Recommends tools - Shows which models work best for which PAL tools

Usage

# Compare current PAL config vs installed models (dry-run)
.\update-pal-config.ps1 -List

# Update PAL config (prompts for confirmation)
.\update-pal-config.ps1

# Prefer -5090 optimized variants (recommended)
.\update-pal-config.ps1 -Prefer5090

# Auto-update without prompting
.\update-pal-config.ps1 -Prefer5090 -NoPrompt

# Specify custom PAL config path
.\update-pal-config.ps1 -PalConfigPath "D:\pal-mcp-server\conf\custom_models.json"

Parameters

Parameter	Description
-List	Compare config vs installed models without changes
-Prefer5090	Only include -5090 variants when both exist
-NoPrompt	Skip confirmation prompts
-PalConfigPath	Custom path to custom_models.json

Example Output

============================================================
  PAL MCP Server Config Updater
============================================================

[INFO] PAL config: C:\Users\...\pal-mcp-server\conf\custom_models.json
[INFO] Found 26 Ollama model(s)

======================================================================
  PAL Config vs Ollama Models Comparison
======================================================================

Models in Ollama NOT in PAL config:
  + deepseek-r1:32b-5090 [5090]
  + qwen2.5-coder:32b-5090 [5090]
  + qwen3:32b-5090 [5090]
  ...

======================================================================
  Recommended Models for PAL Tools
======================================================================

THINKDEEP / CHALLENGE (reasoning, extended thinking):
  * deepseek-r1:32b-5090 (score: 18)
    qwen3:32b-5090 (score: 17)
    qwen3-coder:30b-5090 (score: 17)

CONSENSUS (multi-model debate, diverse perspectives):
  Suggested consensus panel:
    - deepseek-r1:32b-5090 (for)
    - qwen2.5-coder:32b-5090 (against)
    - qwen3:32b-5090 (neutral)

CODEREVIEW / DEBUG / REFACTOR (coding analysis):
  * qwen2.5-coder:32b-5090 (score: 18)
    qwen2.5-coder:14b (score: 18)
    qwen3-coder:30b-5090 (score: 17)

CHAT / QUICK QUERIES (fast responses):
  * llama3.1:8b-5090
  * mistral:7b
  * qwen2.5:3b

* = recommended for this tool category

Model Capabilities Set

The script sets these fields based on model type:

Model Pattern	supports_extended_thinking	intelligence_score
deepseek-r1	✅ true	18
qwen3	✅ true	17
nemotron	✅ true	16
phi4	✅ true	16
qwen2.5-coder	❌ false	18
devstral	❌ false	16-17
llama3.1:8b	❌ false	12
mistral:7b	❌ false	10

Tool Recommendations

PAL Tool	Best For	Recommended Models
thinkdeep	Deep reasoning, complex analysis	deepseek-r1, qwen3, nemotron
challenge	Question assumptions, find flaws	deepseek-r1, qwen3
consensus	Multi-model debate	Mix of reasoning + coding
codereview	Systematic code analysis	qwen2.5-coder, devstral
debug	Root cause investigation	qwen2.5-coder, deepseek-r1
refactor	Code improvement	qwen2.5-coder, qwen3-coder
chat	Quick queries	llama3.1:8b, mistral:7b

Workflow

Typical workflow after installing new models:

# 1. Download new models
ollama pull qwen3-coder:30b

# 2. Optimize for GPU
.\optimize-ollama-5090.ps1 -Model "qwen3-coder:30b"

# 3. Clean up original
.\optimize-ollama-5090.ps1 -Cleanup

# 4. Update PAL config
.\update-pal-config.ps1 -Prefer5090

# 5. Restart Claude Desktop to apply

tip

Use -Prefer5090 to ensure PAL uses your GPU-optimized variants instead of the originals.

limit-ollama-bandwidth.ps1

Limits Ollama download bandwidth to prevent network saturation. Requires Administrator privileges.

Usage

# Interactive menu (prompts for action)
.\limit-ollama-bandwidth.ps1

# Limit to 50% of tested speed
.\limit-ollama-bandwidth.ps1 -Limit

# Limit to 30% of tested speed
.\limit-ollama-bandwidth.ps1 -Limit -Percent 30

# Remove bandwidth limit
.\limit-ollama-bandwidth.ps1 -Unlimit

# Skip speed test, use manual value
.\limit-ollama-bandwidth.ps1 -Limit -SkipSpeedTest -SpeedMbps 100

How It Works

1. Speed Test: Downloads a test file to measure your connection speed
2. Calculate Limit: Computes the target bandwidth (default: 50%)
3. Apply QoS Policy: Creates a Windows QoS policy to throttle ollama.exe
4. Cleanup: Use -Unlimit to remove the policy when done

Parameters

Parameter	Description
-Limit	Enable bandwidth limiting
-Unlimit	Remove bandwidth limiting
-Percent	Percentage of bandwidth to allow (default: 50)
-SkipSpeedTest	Skip speed test, use -SpeedMbps value
-SpeedMbps	Manual speed in Mbps (use with -SkipSpeedTest)

Recommended Workflow

1. Run -Limit before downloading large models
2. Run your setup-ollama.ps1 -AllModels or ollama pull commands
3. Run -Unlimit when downloads are complete

Alternative

If you don't have Admin rights, set the environment variable instead:

$env:OLLAMA_DOWNLOAD_CONN = 1

This limits to a single download connection (fair share with other traffic).

move-ollama-models.ps1

Moves all Ollama models to a different drive and reconfigures Ollama to use the new location. Requires Administrator privileges.

Why Move to SSD?

Model loading is heavily I/O bound. Here's the difference:

Storage	Sequential Read	20GB Model Load
NVMe SSD	3000-7000 MB/s	~7 seconds
SATA SSD	500-550 MB/s	~40 seconds
HDD	80-160 MB/s	~133 seconds

Once loaded into VRAM, disk speed doesn't matter. But if you frequently swap between models, SSD is essential.

Usage

# Move to X:\OllamaModels (default)
.\move-ollama-models.ps1

# Custom target path
.\move-ollama-models.ps1 -TargetPath "D:\AI\Models"

# Just set environment variable (files already moved manually)
.\move-ollama-models.ps1 -SkipMove

Parameters

Parameter	Description
-TargetPath	Destination folder (default: X:\OllamaModels)
-SkipMove	Only set OLLAMA_MODELS env var, don't move files

How It Works

1. Stop Ollama - Stops service and all ollama processes
2. Create Target - Creates destination folder if needed
3. Check Space - Verifies enough free space on target drive
4. Move Files - Uses robocopy /MOVE /MT:8 for reliable, multi-threaded transfer
5. Set Environment - Sets OLLAMA_MODELS system-wide
6. Restart Ollama - Starts service or process
7. Verify - Runs ollama list to confirm

Example Output

=== Ollama Model Migration Script ===

[*] Current models location: C:\Users\You\.ollama\models
[*] Target location: X:\OllamaModels
[*] Free space on X: : 450.23 GB
[*] Current models size: 147.5 GB

[*] Stopping Ollama...
[+] Ollama service stopped

[*] Moving models from C:\Users\You\.ollama\models to X:\OllamaModels...
[*] This may take a while for large models...
[+] Models moved successfully!

[*] Setting OLLAMA_MODELS environment variable...
[+] OLLAMA_MODELS set to: X:\OllamaModels

[*] Starting Ollama...
[+] Ollama service started

=== Migration Complete ===

[+] Models location: X:\OllamaModels
[+] Environment variable OLLAMA_MODELS is set

[*] Verifying with 'ollama list'...
NAME                        ID              SIZE      MODIFIED
deepseek-r1:32b-5090        abc123...       19 GB     2 hours ago
qwen2.5-coder:32b-5090      def456...       19 GB     3 hours ago
...

After Migration

You may need to restart your terminal or log out/in for the environment variable to take effect in all applications.

Disk Space

Ensure your target drive has enough free space. The script checks this before starting, but will fail if space runs out mid-transfer.

scan-ollama-models.ps1

Scans Ollama library for new coder models not in your configuration.

Usage

# Scan for new models with >100 pulls
.\scan-ollama-models.ps1

# Lower threshold to find more models
.\scan-ollama-models.ps1 -MinPulls 50

# Save results to JSON report
.\scan-ollama-models.ps1 -SaveReport

# Check for updates to existing models
.\scan-ollama-models.ps1 -CheckUpdates

How It Works

Since Ollama doesn't have a public browse API, the script maintains a watchlist of known quality model namespaces:

Type	Namespaces
Official	qwen2.5-coder, qwen3-coder, deepseek-coder, codellama, devstral, gpt-oss
Community	NeuralNexusLab/CodeXor, mikepfunk28/deepseekq3, mannix/qwen2.5-coder

The script:

1. Queries each namespace on Ollama's website
2. Extracts pull counts and update dates
3. Compares against your custom_models.json
4. Reports new models above the pull threshold

Parameters

Parameter	Description
-MinPulls	Minimum pull count to consider (default: 100)
-ConfigPath	Path to custom_models.json (default: ./custom_models.json)
-CheckUpdates	Also check if installed models have updates
-SaveReport	Save results to scan-report.json

Example Output

  Checking qwen2.5-coder... 150000 pulls, EXISTS
  Checking NeuralNexusLab/CodeXor... 134 pulls, NEW
  Checking mikepfunk28/deepseekq3... 50 pulls, EXISTS

New models meeting criteria (>= 100 pulls):
  NeuralNexusLab/CodeXor - 134 pulls, updated 2 days ago

To add these models:
  1. Add to $coderModels in setup-ollama.ps1
  2. Add entry to custom_models.json
  3. Run: ollama pull <model-name>

setup-uncensored-models.ps1

Downloads uncensored/unfiltered models for unrestricted conversations.

Usage

# Install uncensored models
.\setup-uncensored-models.ps1

# Force re-download all
.\setup-uncensored-models.ps1 -Force

Models Installed

Model	Author	Focus
dolphin-mistral:7b	Eric Hartford	General uncensored
wizard-vicuna-uncensored:13b	Community	Instruction following
llama2-uncensored:7b	George Sung	Base uncensored
dolphin-phi:2.7b	Eric Hartford	Small uncensored

Use Responsibly

Uncensored models lack safety guardrails. Use for legitimate purposes only.

backup-ollama-models.ps1

Backup and restore models to/from external storage.

Usage

# Show current status
.\backup-ollama-models.ps1 -Info

# Backup to external drive
.\backup-ollama-models.ps1 -Backup -Destination "E:\ollama-backup"

# Restore from backup
.\backup-ollama-models.ps1 -Restore -Source "E:\ollama-backup"

Features

• Robocopy - Reliable large file transfer
• Auto-stop - Stops Ollama before operations
• Space check - Verifies disk space before backup
• Incremental - Only copies changed files

See Backup & Restore Guide for details.

debug-ollama-connection.ps1

Diagnoses and fixes container-to-Ollama connectivity issues.

Usage

# Diagnose only (no changes)
.\debug-ollama-connection.ps1

# Diagnose and fix automatically
.\debug-ollama-connection.ps1 -Fix

# Debug a different container
.\debug-ollama-connection.ps1 -Container perplexica-backend -Fix

Diagnostic Phases

1. Host Check - Is Ollama running on Windows?
2. Container Check - Is container running? Checks restart count, health status, shows logs on issues
3. Connectivity Test - Can container reach Ollama?
4. Gateway Discovery - Auto-detect correct IP for Podman
5. Analysis - Explain the problem
6. Fix - Recreate container with correct settings

Features

• Restart loop detection - Warns if container has restarted >3 times
• Health status check - Reports container health (healthy/unhealthy)
• Auto-show logs - Displays recent logs when issues detected
• Gateway auto-detect - Finds Podman gateway IP automatically

See Podman Troubleshooting for details.

test-ollama-stack.ps1

Comprehensive test suite for verifying your setup.

Usage

# Standard tests
.\test-ollama-stack.ps1

# Include inference tests (slower)
.\test-ollama-stack.ps1 -Full

# Output as JSON (for CI)
.\test-ollama-stack.ps1 -Json

What's Tested

Category	Tests
Prerequisites	Container runtime, NVIDIA GPU
Ollama	API, models, GPU acceleration
Open WebUI	Container health, restart loops, image tag, UI access
Perplexica	SearXNG, backend, frontend (health status for each)

Health Checks

The test suite detects:

• Restart loops - Fails if container has restarted >3 times
• Health status - Reports healthy/unhealthy/starting
• Auto-show logs - Displays recent logs on failures

Example Output

========================================
  Ollama Stack Test Suite
========================================

[Prerequisites]
  [PASS] Podman is running
  [PASS] NVIDIA GPU detected (GeForce RTX 5090)

[Ollama]
  [PASS] API responding on :11434
  [PASS] 6 models available
  [SKIP] No models currently loaded
  [SKIP] Inference test (use -Full flag)

[Open WebUI]
  [PASS] Container running (health: healthy)
  [PASS] Using CUDA image (no embedded Ollama)
  [PASS] UI accessible on :3000

[Perplexica]
  [PASS] SearXNG container running (health: healthy)
  [PASS] SearXNG accessible on :4000
  [PASS] Backend container running
  [PASS] Frontend container running
  [PASS] Frontend accessible on :3002

========================================
  Results: 12 passed, 0 failed, 2 skipped
========================================

sync-container-images.ps1

Mirrors container images to your own GitHub Container Registry.

Usage

# Check for updates (no changes)
.\sync-container-images.ps1

# Sync changed images
.\sync-container-images.ps1 -Sync

# Force sync all
.\sync-container-images.ps1 -Force

# Sync specific image
.\sync-container-images.ps1 -Sync -Image "open-webui:cuda"

Images Tracked

Image	Upstream
open-webui:cuda	ghcr.io/open-webui/open-webui:cuda
open-webui:main	ghcr.io/open-webui/open-webui:main
searxng:latest	docker.io/searxng/searxng:latest
perplexica-backend:main	docker.io/itzcrazykns1337/perplexica-backend:main
perplexica-frontend:main	docker.io/itzcrazykns1337/perplexica-frontend:main

See Container Sync for details.

cleanup-containers.ps1

Stop and remove all Ollama stack containers.

Usage

# Interactive - asks what to delete
.\cleanup-containers.ps1

# Remove containers only (keep data)
.\cleanup-containers.ps1 -Force

# Remove containers AND volumes (deletes all data)
.\cleanup-containers.ps1 -DeleteVolumes

# Remove everything, no prompts
.\cleanup-containers.ps1 -DeleteVolumes -Force

What It Removes

Container	Data Location
open-webui	open-webui volume
perplexica-frontend	-
perplexica-backend	./perplexica/data
searxng	./searxng

Options

Parameter	Description
-Force	Skip confirmation prompts
-DeleteVolumes	Also remove volumes (persistent data)

Common Parameters

Verbose Output

Most scripts support verbose logging:

.\setup-ollama.ps1 -Verbose

What-If Mode

Some scripts support dry-run:

.\backup-ollama-models.ps1 -Backup -WhatIf

Help

All scripts include help:

Get-Help .\setup-ollama.ps1 -Full

Or use the comment-based help at the top of each script.