Ollama Integration

DSO uses Ollama for local AI-powered analysis. This guide covers setup, configuration, and troubleshooting.

What is Ollama?

Ollama is a tool for running large language models locally on your machine. DSO uses it to:

• Analyze security scan results intelligently
• Explain vulnerabilities in natural language
• Prioritize issues based on business impact
• Identify false positives

Installation

macOS

brew install ollama

Linux

curl -fsSL https://ollama.ai/install.sh | sh

Windows

Download from ollama.ai or:

winget install Ollama.Ollama
# or
scoop install ollama

Starting Ollama

macOS / Linux

ollama serve

This starts Ollama in the foreground. For background:

# macOS (using launchd)
brew services start ollama

# Linux (using systemd)
systemctl --user start ollama

Windows

Ollama runs as a service automatically after installation.

Downloading Models

DSO works with various models. The installation script automatically installs qwen2.5:7b by default.

Default: qwen2.5:7b

ollama pull qwen2.5:7b

• Size: ~4.7 GB
• Speed: Fast
• Quality: Excellent for security analysis
• Default: Automatically installed if no models exist

Alternative: llama3.1:8b

ollama pull llama3.1:8b

• Size: ~4.7 GB
• Speed: Fast
• Quality: Excellent for security analysis

Lightweight: phi3

ollama pull phi3

• Size: ~2.3 GB
• Speed: Very fast
• Quality: Good for basic analysis

Other Options

ollama pull mistral:7b # ~4.1 GB - Good balance
ollama pull gemma:7b # ~5.2 GB - Google's model
ollama pull llama3.1:70b # ~40 GB - Best quality (requires more RAM)

Configuration

Model Selection

DSO uses the following priority order:

1. Environment variable DSO_MODEL
2. Configuration file ~/.dso/config
3. Default: qwen2.5:7b

Via Environment Variable

# macOS / Linux
export DSO_MODEL=phi3
dso audit .

# Windows (PowerShell)
$env:DSO_MODEL="phi3"
dso audit .

Via Configuration File

The installation script automatically saves the default model to ~/.dso/config:

# View current config
cat ~/.dso/config
# Output: DSO_MODEL=qwen2.5:7b

# Change model
echo "DSO_MODEL=llama3.1:8b" > ~/.dso/config
dso audit .

Interactive Model Selection

During installation, you can select models interactively. The selected model is saved to ~/.dso/config.

Remote Ollama

If Ollama is on a different machine:

export OLLAMA_HOST=http://192.168.1.100:11434
dso audit .

Permanent Configuration

Add to your shell profile:

macOS / Linux (~/.zshrc or ~/.bashrc):

export DSO_MODEL=qwen2.5:7b # Default model
export OLLAMA_HOST=http://localhost:11434

Windows (PowerShell profile):

$env:DSO_MODEL="qwen2.5:7b"
$env:OLLAMA_HOST="http://localhost:11434"

Or use configuration file (recommended):

# ~/.dso/config
DSO_MODEL=qwen2.5:7b

The configuration file is automatically created during installation.

Verification

Check that everything works:

dso check

This verifies:

• Ollama connection
• Available models
• Configured model is installed

How DSO Uses Ollama

1. Analysis

When you run dso audit, DSO:

1. Scans your codebase
2. Formats results for AI
3. Sends to Ollama with a security-focused prompt
4. Receives intelligent analysis
5. Displays prioritized findings

2. Explanation

When you run dso why, DSO:

1. Sends vulnerability details to Ollama
2. Receives natural language explanation
3. Displays context-aware analysis

Performance Tips

Model Selection

• Development: Use phi3 for faster feedback
• Production: Use qwen2.5:7b (default) for excellent analysis
• Large Projects: Use llama3.1:70b for best quality (requires ~40 GB RAM)

Resource Usage

Models use RAM:

• phi3: ~2.3 GB RAM
• qwen2.5:7b: ~4.7 GB RAM (default)
• llama3.1:8b: ~4.7 GB RAM
• mistral:7b: ~4.1 GB RAM
• gemma:7b: ~5.2 GB RAM
• llama3.1:70b: ~40 GB RAM

Ensure you have enough RAM available.

First Run

The first analysis may be slower as the model loads. Subsequent analyses are faster.

Troubleshooting

Ollama Not Responding

Problem: connection refused

Solutions:

1. Start Ollama: ollama serve
2. Check if running: ollama list
3. Verify port: Default is 11434
4. Check firewall settings

Model Not Found

Problem: model not found

Solution: Download the model:

ollama pull llama3.1:8b

Slow Responses

Problem: AI analysis is slow

Solutions:

1. Use a lighter model: export DSO_MODEL=phi3
2. Ensure enough RAM available
3. Close other applications
4. Use a GPU if available (Ollama supports GPU acceleration)

Out of Memory

Problem: Model doesn't fit in RAM

Solutions:

1. Use a smaller model: phi3
2. Increase available RAM
3. Close other applications

Advanced Configuration

Custom Prompts

You can customize the AI prompt by creating:

templates/prompt_system.txt

Or in your home directory:

~/.dso/templates/prompt_system.txt

Streaming Responses

DSO uses streaming for real-time output. This is handled automatically.

Context Management

DSO maintains conversation context for better analysis. This is handled automatically.

Integration with CI/CD

For CI/CD, you can:

1. Install Ollama in CI:

- name: Install Ollama
run: curl -fsSL https://ollama.ai/install.sh | sh

2. Download Model:

- name: Pull model
run: ollama pull qwen2.5:7b

3. Run DSO:

- name: Run audit
run: dso audit . --format json

Note: CI environments may have limited resources. Consider using smaller models or running without AI analysis.

About the Developer

DSO is developed by Ismail MOUYAHADA, a DevSecOps Engineer and Multi-Platform Software Developer.

With expertise in security automation, cloud infrastructure, and cross-platform development, Ismail created DSO to bring enterprise-grade security analysis to developers' terminals with zero configuration and complete privacy.

See Also

• check: Verify Ollama setup
• audit: Run security audit with AI
• why: Get AI explanations
• Ollama Documentation